mailto:support@psi.cz
Photon Systems InstrumentsPhoton Systems Instruments
User Guide for PSI models including: Morpho PlantScreen Analyzer, PlantScreen Analyzer, Analyzer
| PSI
File Info : application/pdf, 21 Pages, 2.85MB
DocumentDocument Manual Version: 2024/02 © PSI (Photon Systems Instruments), spol. s r.o. www.psi.cz
This document and its parts can be copied or provided to a third party only with the express permission of PSI. The contents of this manual have been verified to correspond to the specifications of the device. However, deviations cannot be ruled out. Therefore, a complete correspondence between the manual and the real device cannot be guaranteed. The information in this manual is regularly checked, and corrections may be made in subsequent versions. The visualizations shown in this manual are only illustrative. This manual is an integral part of the purchase and delivery of equipment and its accessories and both Parties must abide by it.
TABLE OF CONTENTS
1
Introduction .......................................................................................................................................... 5
1.1
Purpose .........................................................................................................................................................5
1.2 2 2.1 2.2 2.2.1 2.2.2 2.3 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5 2.4 2.4.1 2.4.2 2.4.3 2.4.4 2.4.5 2.5 2.6 3 4
Layout ...........................................................................................................................................................5 Morpho Analyzer................................................................................................................................... 6 Data Source Folder .......................................................................................................................................6 Mask Builder .................................................................................................................................................6 Use of Tray Mask ..........................................................................................................................................6 Drawing Tray Mask .......................................................................................................................................6 Plant Mask Tool ............................................................................................................................................7 Input..............................................................................................................................................................7 Color Picker Tool ...........................................................................................................................................8 Mask Detection Tools ...................................................................................................................................9 Plant mask settings .....................................................................................................................................13 Masking ....................................................................................................................................................... 15 Settings .......................................................................................................................................................16 Preprocessing Options ................................................................................................................................16 Plant Mask Settings.....................................................................................................................................16 Tray Mask....................................................................................................................................................16 Computed Parameters................................................................................................................................17 Color Segmentation ....................................................................................................................................18 Single Run ...................................................................................................................................................18 Batch Run....................................................................................................................................................19 Bug Reporting...................................................................................................................................... 21 List of Figures ...................................................................................................................................... 21
Page | 3
Page | 4
1 INTRODUCTION
1.1 PURPOSE
The Morpho Analyzer application is designed for users, who need to run morphological and color analysis of plant images stored locally on a computer disc drive. Typical usage is to run an additional analysis or test analysis on plant images exported to a local storage from PlantScreenTM System database. This manual describes Morpho Analyzer version 1.0.14.3.
1.2 LAYOUT
The application window is based on a horizontal tabbed menu, which displays each tab across the whole window when clicked. The tabbed menu is arranged from left to right as the analysis steps gradually process. The initial windows Mask Builder, Plant Mask Tool and Settings are used for analysis settings. The final windows Single Run and Batch Run are used to analyze images based on settings defined in the Settings window.
Page | 5
2 MORPHO ANALYZER
2.1 DATA SOURCE FOLDER
The Morpho Analyzer application is designed to process locally stored data, which must be kept in a separate folder. When obtaining plant images with automatic measurement using PlantScreen systems, all data is stored in a database. To create a local folder containing images or mask files obtained with PlantScreen system use the export part of PlantScreen Data Analyzer application. The images are exported only in PNG format, which is currently the only image format supported with MorphoAnalyzer. The file format used for tray mask and plant mask is XML, which is also compatible with Morpho Analyzer.
2.2 MASK BUILDER
2.2.1 USE OF TRAY MASK
The Mask Builder window Fig. 1. allows to draw one or more areas subjected to detection of plant specific pixels. To identify the plants, each area is marked with a unique name. Typically, one area is used for detection of a single plant individual. One or more areas designated for the detection of plant masks are referred to as "tray mask". Before using the tray mask, it's essential to load a template background image onto the viewing panel. (Fig. 1-10). Click the Load Template button (Fig. 1-1) and select an image to load from folder. Then, optionally a tray mask can be loaded on the image by clicking the Load Mask (Fig. 1-2) button. Once loaded or drawn, the mask can be saved to a local storage with Save Mask button (Fig. 1-3). Load the current mask to Settings window with Use Mask button (Fig. 1-4) or to the Plant Mask Tool with Use for Plant Mask Tool button (Fig. 1-5).
Fig. 1 Mask Builder Tab
2.2.2 DRAWING TRAY MASK
Drawing mode is activated by selecting an area shape button (Fig. 1-6) and marked with a red ribbon on top of the view panel. After choosing a suitable shape type by clicking rectangle , ellipse , pie or polygon button, the shape is drawn by dragging the mouse while holding left-click. To select multiple plants on the image, it is necessary to create multiple nonoverlapping areas, as is shown in Fig. 1. Polygon shape is drawn differently. Selection of the polygon button activates a mode, where mouse left-clicking adds single points to a newly created shape. When the shape contains at least three points, the drawing can be finished with mouse right-click connecting the last point with the first. If less than three points are present, the right click discards the shape and the drawing mode is canceled.
Page | 6
The drawing mode is automatically deactivated after a shape is drawn. After deactivation, mouse left-click selects an area and when holding the Shift key, mouse left-click on another area selects multiple areas. If multiple areas are selected, each of the following operation is applied to all items. To move area(s), place the cursor on the inner part of selected area and hold mouse-left click while dragging the mouse. To resize the area, place the cursor on the border line and hold mouse-left click while moving the mouse. To achieve symmetrical shape of a square or a circle, press Ctrl key while resizing. Another tools can be activated with the following buttons (Fig. 1-
7): copy selection to clipboard, zoom in the image.
paste clipboard content on image,
delete selection,
toggle displaying area names,
All drawn areas are listed in the Areas tab (Fig. 1-9). To select an area, left click the item in the list. Multiple areas can be selected by clicking the items individually. The listed areas are shown in the view panel (Fig. 1-10), either with red border line when selected, or green border line when not selected. The Selected Item panel (Fig. 1-8) can be used to edit area name or position by typing new values to textboxes. All changes must be confirmed with Enter key.
When multiple areas are selected, only parameters common for all selected areas are displayed in the Selected Item tab.
Area name must be unique within the scope of the mask. If the condition is not met, duplicate names are selected with warning dialog after clicking Save Mask or Use Mask button.
2.3 PLANT MASK TOOL
The tray mask contains definition of one or more areas, which are used for plant masking. This involves the identification of individual or multiple plants present on a tray. The identification of a plant-masked object within a designated area of the tray mask is achieved either through a pixel-by-pixel evaluation of color formulas or a binary classification of each pixel using random forest model. Additionally, other optional techniques, which re-select and regroup the selected pixels within an area, can be selected. Users have the flexibility to employ either the color formula method or the random forest method in the Plant Mask Tool tab.
2.3.1 INPUT
To start using Plant Mask Tool, load sample images and then a tray mask to the viewing panel (Fig. 2-4) by clicking the Load Images and Load Tray Mask (Fig. 2-1) buttons. Sample images: Within the Plant Mask Tool tab, users can upload any number of images. However, it's important to be mindful that the processing time and computational costs increase with each additional image. By clicking the Load Images button, any presently open images are automatically closed and new images can be selected from a designated folder. To display a specific image in the viewing panel, simply click on its corresponding name within the image bar. To navigate among images that are currently not visible, employ the arrows for switching between their names (Fig. 2-3). The presently visualized image is highlighted with bold letters within the image bar for easy identification. In the Tab options section (Fig. 2-2), there is the Add image button which allows adding more images to the existing selection. Additionally, the Close current tab button is available to swiftly shut the current image display. If you intend to close all loaded images at once, simply press the Close all button. Tray mask: The tray mask can be loaded from the file using Load Tray mask button. The tray mask can be also loaded from the Mask Builder window - Mask builder tab Use for Plant Mask Tool button (see the previous chapter). The same tray mask is used for all uploaded images. To improve visibility, re-design the border line of the tray mask. In Masking panel (Fig. 2-9), the number represents actual thickness of the line and the colored rectangle on the right represents actual color of the line. To change the color, Color Selection window pops-up after right-clicking the colored rectangle.
Page | 7
Fig. 2 Plant Mask Tool Tab Before Color Segmentation
2.3.2 COLOR PICKER TOOL
When the image and the tray mask are loaded and displayed, begin by creating a color dataset comprising both wanted colors, representing plant pixels, and unwanted colors, representing background pixels. Activate color picker tool by flagging either the Picking mode or Drawing mode checkbox (Fig. 3-1). Then, select several color points (i.e. pixels) or draw one or more shapes by mouse leftclicking or moving mouse with left-click on the area to add colors to corresponding Wanted Colors or Unwanted Colors tab list (Fig. 2-5, Fig. 2-8). In either mode of selection, it is necessary to carefully select only pixels corresponding to the plant in Wanted colors panel and only pixels corresponding to the background inside the area in Unwanted colors panel. When selecting colors, it can be useful to activate the zoom tool (Fig. 2-11) by unflagging the checkbox Fit. Move the slider from left to right across prescribed positions corresponding to 50, 100, 200 or 400 % zoom. Use scroll bars located in the view panel (Fig. 2-4) to navigate around the image.
Fig. 3 Color picker tool
Page | 8
Creating the datasets is a crucial step. A high number of selected colors increases the computation time for both segmentation methods. For Color Formula Computation, using K-means clustering is recommended to reduce the dataset size. However, this approach is not suitable for the Random Forest Method because K-means clustering does not preserve the pixel coordinates, which are essential for its functionality. For the Random Forest Method, it is advisable to use datasets containing up to 500,000 pixels to train a model effectively. Right-clicking on the table of selected colors allows users to delete the currently selected pixel from the dataset (this can also be done by pressing the delete key), clear the list of colors, or apply K-means clustering to reduce the number of colors. To use K-means clustering, hover the cursor over the Apply K-means option and select the desired number of color clusters from the drop-down menu using the left mouse button. (Fig. 3-2). Before you begin to formulate any segmentation strategy, it's important to consider the balance between your datasets of wanted and unwanted colors. It is advisable to ensure that these datasets are of similar size (balanced) to facilitate optimal segmentation performance. When attempting segmentation, the software evaluates the balance between the datasets of wanted and unwanted colors by calculating the ratio of their sizes. If this ratio falls within the range of 0.7 to 1.3, the datasets are considered balanced, and segmentation proceeds. If the ratio falls outside this range, a warning message is displayed to indicate that the datasets are not balanced and may affect segmentation quality.
2.3.3 MASK DETECTION TOOLS
The Mask Detection Tools (Fig. 2-6) include two methods for plant mask creation, accessible from a horizontal tab menu (Fig. 4-1). The first tab page, labeled RGB Formula Computation, employs a basic thresholding segmentation method. It is used to generate multiple color formulas, calculate optimum threshold and score precision for any formula of interest. RGB formula computation To quickly compute a set of color formula definitions, click on Suggest Formula Faster. For more accurate computation with higher precision score, use Suggest Formula Better button (Fig. 4-2). Each formula name in the list of suggested color formula definitions contains bracketed addition behind the formula, which marks whether the image was pre-processed for better performance with reflection reduction (RR). If only standard color formula computation was used (i.e. without RR), then the item is marked with (std) addition (Fig. 4-3).
Fig. 4 RGB Formula Computation tab page in Mask Detection Tools After selecting a color formula by left-mouse click, add the formula to the formula box (Fig. 4-4) by right-clicking anywhere on the table to show a drop-down menu with the Add Selected Formula option. If the formula uses reflection reduction, the checkbox Use reflection reduction is flagged automatically. The Show option in the drop-down menu can be used to adjust maximum number of formulas shown in the list. To remove a single formula item, click the minus (-) button. It is also possible to type a new formula definition directly to the upper text-box and add it to the formula box with plus (+) button (Fig. 4-4) The optimum threshold for each color formula is computed by Auto threshold tool. After adding the formula to the formula-box, click the button Compute to gain the threshold line for standard and for reduced reflection conditions. The Show Graph button (Fig. 4-5) opens graphs of individual computed color values divided by the threshold line (Fig. 5).
Page | 9
Fig. 5 Graph display performance of the selected threshold line To load the selected color formula with threshold to the Plant Mask settings tab (Fig. 2-7), click the Use selected formula button located on the bottom of the Mask Detection Tools (Fig. 4-6) panel. Machine learning computation The second page tab in the Mask Detection Tools section, Machine Learning Computation (Fig. 6), adopts a more advanced approach by using a machine learning model to analyze more complex images. This tab features pixel binary classification through the use of a random forest model.
Fig. 6 Machine Learning Computation tab page in Mask Detection Tools This method employs four preprocessing approaches, three of which can be adjusted by user. Upon clicking the Preprocessing Options button (Fig. 6-1), a new window appears (Fig. 7), where the size of these three features can be selected. The purpose of each feature is detailed in Tab. 1. All three incorporate matrices with sizes that can be modified in the Preprocessing Options window (Fig. 7-1, Fig. 7-2, Fig. 7-3). After choosing the preferred sizes, confirm your selections by clicking the Ok button (Fig. 7-4). These approaches, when integrated as features in the random forest model, not only enhance its performance but also significantly improve the quality of segmentation by enabling more accurate differentiation between image regions. This leads to more precise and valuable segmentation outcomes.
Page | 10
Fig. 7 Preprocessing Options for machine learning plant mask computation
Tab. 1 Brief description of Preprocessing Options representing Random Forest model features
Feature name
Description
Gauss Filter
The Gauss Filter, also known as the Gaussian blur, is a smoothing technique used in image processing to reduce noise. This results in a softening effect, making it useful for blurring images, enhancing image quality. Valid range: 4 25. Set to 0 and this feature won't be used.
Sobel Operator
Sobel Operator employs edge detection, a gradient-based method used in image processing to identify the edges of objects within an image by detecting changes in brightness. Valid range: 3 15. Set to 0 and this feature won't be used.
Entropy
In the context of image texture analysis, entropy measures the randomness or complexity within an image's texture, quantifying the amount of information or variability in the intensity patterns. This metric is used to differentiate between areas of an image with varying levels of detail or texture, with higher entropy values indicating more complex, information-rich textures. Valid range: 1 25. Set to 0 and this feature won't be used.
Histogram Equalization
Histogram equalization is a technique to adjust the contrast of images, enhancing features by evenly distributing the intensity levels. This preprocessing step can improve the model's ability to recognize patterns and distinctions within the data.
This feature is created from the image blurred with Gauss filter, therefore depends on its values range.
Examples of different size values of Preprocessing Options: Gauss Filter
Size = 25 (max. value)
Sobel Operator Size = 15 (max. value)
Page | 11
Gauss Filter Size = 7 (default value)
Sobel Operator Size = 7 (default value)
Size = 4 (min. value)
Size = 3 (min. value)
Entropy Size = 25 (max. value)
Histogram Equalization Size = 25 (max. value)
Page | 12
Entropy Size = 7 (default value)
Histogram Equalization Size = 7 (default value)
Size = 1 (min. value)
Size = 4 (min. value)
After selecting the Preprocessing Options, the model's training is initiated by clicking the Train Model button (Fig. 6-2). The computation time for this operation depends on the number of loaded images, tray mask areas, and the size of the color dataset. The training uses 80% of the dataset, while the remaining 20% is reserved for model evaluation. The results of this evaluation are displayed in a table (Fig. 6-3), which includes the model name, its accuracy (the proportion of true results among the total number of cases examined), AUC (Area Under the ROC Curve, measuring the model's ability to distinguish between classes), F1 score (a harmonic mean of precision and recall, indicating the balance between the model's accuracy and its completeness), and time it took to train the model in milliseconds. The name of the trained model appears in the drop-down list below, from which you can select the Chosen Model to be used for segmentation. This list also includes models added when you click the Load Model button (Fig. 6-4). Trained models can be saved as a ZIP file to any directory by clicking the Save Model button (Fig. 6-5). To carry out the segmentation and generate the plant mask, click the Compute Plant Mask With Chosen Model button (Fig. 6-6). This action applies the model selected from the drop-down list, referred to as the Chosen Model, to all pixels within the tray mask areas, thereby creating the Plant Mask. The model name is then transferred to the Plant Mask settings section, where additional postprocessing and filtering are applied to the created Plant Mask.
2.3.4 PLANT MASK SETTINGS
Image analysis can be optionally enhanced with other techniques displayed in Plant Mask Settings panel. The settings can be saved or loaded with Load or Save buttons located on the bottom of the Plant Mask Settings tab (Fig.8-2). Depending on the segmentation method chosen, there are two configurations for Plant Mask settings. For the basic method, known as RGB Formula Computation, the settings are visually depicted in Fig. 8, while for the more advanced Machine Learning Computation method, the settings are shown in Fig. 9. A brief description of all the parameters available for each method can be found in Table 2.
Page | 13
Tab. 2 Summary of Image Analysis parameters
Parameter name Formula Threshold Median Filter Size Min Size
RGB Formula Computation Description Color formula definition. Use whole numbers (1, 2, 3, etc.), operators (+, -, *, /, min, max, ln, log, sqrt, ^) and pixel color intensity variables (R, G, B). Thresholding the computed color formula value is used to distinguish between plant containing pixels (above threshold) and the background (below threshold). Median 2D filter is used to reduce noise by filtering with pixel window n x n. When plants are clearly visible, higher n values can be used to increase efficiency. Set value of minimal size of a single plant mask selected area [px].
Min Hole Size
Crop Object on Borders
CropObjectsOnBordersMin PreserveSize [px]
Minimal size of holes in the mask objects in pixels, typically tens of pixels. The holes smaller than this value are closed and taken into the object pixels. [px]
Flag the checkbox to crop protrusions or unrelated objects (e.g. leaves of other plants) appearing on the borders of the tray mask.
Minimum size of objects that will not be removed even if they are on the edge of the area specified by the tray mask (objects smaller than this value will be cropped on the borders if CropObjectsOnBorders option is set to true)
Use Reflection Reduction Skip Bad Exposed Points
Option normalizing RGB values in each pixel (ratio between channels is then considered for thresholding rather than absolute pixel values). Useful for harsh imaging conditions e.g. light reflection from petri dishes.
Flag the checkbox to crop poorly exposed pixels from the plant mask. Poorly exposed pixels have at least one color at minimum or maximum value.
Machine Learning Computation
Parameter name Chosen Model
Set Decision Threshold
Description Name of a model selected for segmentation; trained or loaded. The random forest model assigns each pixel a certain probability of belonging to one of two categories: plant or background. The default threshold for classification is set at 0.5, but this can be adjusted.
Dilation Erosion
Increases the size of the plant areas by applying a disk of a specific, adjustable size; potentially connecting nearby plants and filling in small gaps within the plant shapes.
Reduces the size of the plant areas in a plant mask by using a disk of a specific, adjustable size; helping to remove small, isolated patches mistakenly identified as plants and sharpening the boundaries between plants and the background.
Opening Closing
Conducted by first applying erosion and then dilation with a disk of a specific, adjustable size; separates closely situated plants in a plant mask by removing small connections between them and eliminating tiny specks or noise, clarifying the individual plant shapes.
Performed by first applying dilation followed by erosion using a disk of a specific, adjustable size; smooths the edges of plant shapes in a plant mask, closes small holes, and connects close parts, making the plant areas appear more cohesive.
Page | 14
Fig. 8 Plant mask settings for RGB Formula Segmentation
Fig. 9 Plant mask settings for Machine Learning Computation
2.3.5 MASKING
Clicking the Recompute Plant Mask button (Fig. 2-8, Fig. 8-3) in case of RGB Formula Computation generates Plant Mask. The color formula and optimum threshold is computed for each pixel of the image and other optional image analysis techniques listed in Plant Mask Settings are applied. On the other hand, in case of Machine Learning Computation the Recompute Plant Mask button (Fig. 9-3) serves to filter the existing Plant Mask according to the parameters set in Plant Mask Settings. Then, the view panel (Fig. 11-2) is refreshed with the resulting masked image. The image can be viewed with plant mask border line placed on top of the image and/or in combination with tray mask border line and with or without background. The view options are defined by flagging checkboxes in the Masking tab (Fig. 10). The border lines can be restyled by thickness and color options. The thickness is determined by the number in the text-box and the color can be set by clicking the colored rectangle and selecting a color from Color pop-up window.
Fig. 10 Masking
If the analysis yields unsatisfactory results, consider adding new colors to the existing list in color picker tool (Fig. 111) and computing a new set of formulas, or train a new model. Alternatively, you can select a different configuration of filtering parameters in Plant Mask Settings (Fig. 8).
Page | 15
Fig. 11 Plant Mask Tool Tab After Color Segmentation
In the end, the settings, which yield most accurate masking result are sent to the Settings window by clicking the Use Plant Mask Settings (Fig. 11-3) button. It is recommended to also save the file separately to a local storage for later use or reference (Fig. 8-3, Fig. 9-3, button Save).
2.4 SETTINGS
The Settings window displays final analysis settings, which is then used in Single Run and Batch Run windows. The application stores the settings automatically when closed and restores them after restart.
2.4.1 PREPROCESSING OPTIONS
If the advanced Machine Learning Computation method is selected in the Plant Mask Tools tab, the Preprocessing Options settings (Fig. 12-1) become visible. Here, the settings are passed through and can be changed.
2.4.2 PLANT MASK SETTINGS
Plant Mask Settings tab (Fig. 12-2) contains correction settings for barrel lens distortion and plant mask detection. To correct for barrel (fish-eye) distortion, specify a correction degree with a number from -1 to 1. The checkbox x- fixed or y- fixed omits correction in x or y axis when flagged, which is useful for correcting images acquired in line-scan mode where distortion occurs only in one direction. When check-boxes for both axis are flagged, none correction is done. Centering of image correction can be done after flagging the Change fish -eye center checkbox and filling new values of x- or y- axis.
2.4.3 TRAY MASK
The button Load Tray Mask in Tray Mask Panel (Fig. 12-3) is used to select and load a tray mask for use in analysis. In gradually processed analysis steps starting in Mask Builder window, the tray mask should be already added to the Settings window, which is indicated by green colored Load Tray Mask button. If the button remains gray, the tray mask must be loaded by clicking the Load Tray Mask button and selecting the mask file in Open window. The tray mask can be also added with the Use Mask button (Fig. 1-4) in the Mask Builder window. Then, preview the mask on a selected image after clicking the Show Tray Mask button (Fig. 12-3).
Page | 16
Fig. 12 Settings Tab
2.4.4 COMPUTED PARAMETERS
Morphology parameters are selected in the Computed Parameters panel (Fig. 12-4). Clicking the button Side View or Top View Parameters automatically flags a set of relevant parameters. All morphology parameters are listed and briefly described in Tab. 3.
Tab. 3 Morphology Parameter Description
Parameter
Description
AREA_PX PERIMETER_PX ROUNDNESS Formula: ROUNDNESS2 Formula: ISOTROPY
Formula: COMPACTNESS
Formula: ECCENTRICITY
Formula: RMS
Formula: SOL Formula: WIDTH_PX
Total area covered with plant. Unit: pixel Length of plant perimeter. Unit: pixel Ratio between area and perimeter of plant surface. 4 * PI * Area / Perimeter² Ratio between area and perimeter of plant convex hull. 4 * PI * Convex_Hull_Area / Convex_Hull_Perimeter² Isotropy of the object, computed as ratio of area and perimeter of the polygon created by object vertices. 4 * PI * Polygon_Area / Polygon_Perimeter²
Ratio between area and surface of convex hull enveloping particular plant. More "compact" plants usually have shorter petioles and massive leaf blades. Area / Convex_Hull_Area
Parameter describing the degree of difference between convex hull area and circle, which has center in plant centroid. {Area (circle only) + Area (convex hull only)} / Area (intersection)
Rotational Mass Symmetry, based on the fitting of the ellipsis with identical second central moment as the analyzed object. 2 * {sqrt ((1/2 * Major_Axis_Length)² - (1/2 * Minor_Axis_Length)²)} / Major_Axis_Length
Slenderness of Leaves. Skeleton Perimeter² / Area Width of bounding box enveloping plant. Unit: pixel
Page | 17
HEIGHT_PX
Heigth of bounding box enveloping plant. Unit: pixel
2.4.5 COLOR SEGMENTATION
After masking the image, color segmentation can be used to classify the pixel-by-pixel color representation of the masked object. In color segmentation, the number of colors is reduced by k-means clustering technique, which classifies colors based on similarity. Each class shares properties similar to a color defined as representative. The Color Definitions list in the Color Segmentation tab (Fig. 12-5) is used to define representative colors. The Add and Delete buttons allows to add or remove one selected color definition to/from the list. You can edit the RGB values for any representative color in the list by selecting and double-clicking the item. From the Color pop-up window select a definition from the Basic colors part or define a custom color in the Custom colors part. The Color generating tab (Fig. 12-6) is a tool for automatic generation of color classes of plant(s) masked according to the settings definition in the Setting window (Fig. 12-1 and Fig. 12-2). To set the number of clusters for the K-means algorithm, you can either specify the number manually or select the Auto-Detect Cluster Count checkbox. The software will then employ the Silhouette method to assess clustering effectiveness with 2 to 20 clusters, choosing the number that yields the highest score. Then click Generate colors button. In the Open file pop-up window select a folder of source images used for masking and automatic color definition. The progress bars in Color generating tab (Fig. 13-1) show how the color generator progresses. When done Finished appears on the upper bar (Fig. 13-2).
Fig. 13 Color generating in progress (1) and finished (2)
Color definitions can be exported as a TXT file with the Store Definitions button or loaded with the Load Definitions button in the Color Segmentation panel (Fig. 12-5). The exported file containing the color hue definitions, can subsequently be imported into the Local Analysis section of the PlantData Analyzer software. This allows for the use of the generated color channels to run color segmentation on data stored in the database. The result of the color segmentation analysis is formulated as a number of pixels in each color class for each plant mask selected object.
2.5 SINGLE RUN
After settings up analysis in the Settings tab, click Show Tray Mask button or select Single Run window in the horizontal tabbed menu. The Single Run window (Fig. 14) is used to analyze and preview a single sample image to verify analysis settings.
Page | 18
Fig. 14 Single Run Tab In the tab Single Run window load a sample image with the Load Image button (Fig. 14-1). Click the button Run Analysis (Fig. 14-2) to start analysis using the basic RGB Formula Computation. If you want to run the advanced Machine Learning Computation, click the Run ML Analysis button (Fig. 14-3). After completion of analysis, a table with calculated results appear in the table view panel (Fig. 14-4).
The values shown in the result table can be selected by clicking the table panel and selecting required cells. Then, the cells can be copied to clipboard and inserted to a text or spreadsheet editor. The resulting masked image is displayed in the view panel (Fig. 14-6). In Show Image panel (Fig. 14-5) pick one of three viewing modes: original, masked and color segmented to visualize the masking and color segmentation result. To save the plant mask or the image click Save Plant Mask or Save Image button (Fig. 14-7).
2.6 BATCH RUN
The Batch Run tab (Fig. 15) is used for analysis on a set of images stored in a local storage folder.
Fig. 15 Batch Run Tab
Page | 19
After verifying correct analysis settings in the Single Run tab, load a whole set of input images to the Batch Run tab. Click the Select Images Folder button (Fig. 15-1) and select the image folder in the Browse for Folder pop-up window. In the Storing Path text-box (Fig. 15-2) define a path for storing analysis result. Click Select Path button (Fig. 15-2) to create and select the folder in the Browse for Folder pop-up window. Then, click Run Analysis (Fig. 15-3) and the basic RGB Formula Computation image analysis starts. If you want to run the advanced Machine Learning Computation, click the Run ML Analysis button (Fig. 15-4). During analysis, the result is continuously updating on the table view panel (Fig. 15-5) and on the image view panel (Fig. 15-6), the last processed image is shown. Analysis results are being exported to the output folder defined in the Storing Path (Fig. 15-2). The folder contains a result table, the used plant mask and three sets of images: fish-eye corrected, masked and color segmented. · results.csv: This file contains comma separated values of parameters defined in analysis settings. The file format is compatible with
most text spreadsheeed editors. · PlantMask_originalfilename.xsel: Plant mask definition file.
FEC_originalfilename.png Original image corrected for fish-eye distortion.
WithMask_originalfilename.png: Fish-eye corrected image with cropped plant masked objects.
Coloring_originalfilename.png Image with masked plants colored based on the result of color segmentation.
In addition to shape definition, tray masks used in PlantScreen systems also contain information about pixel size. When pixel size value defined in the tray mask is not equal to zero, several morphological parameters with pixel units are also calculated with milimeter units. These parameters, such as Area (mm2), Perimeter(mm), Width (mm) and Height (mm), are automatically included in the results table (Tab. 4).
Page | 20
Tab. 4 Results Table
3 BUG REPORTING
Even though we are extensively testing each released version of the software, some bugs still may pass our testing process. Please consider reporting bugs to make this software better. If you encounter a bug or application crash please send a mail with "Morpho
Analyzer bug" text in the subject to support@psi.cz. In the message body, describe the sequence of the steps leading to the error.
4 LIST OF FIGURES
Fig. 1 Mask Builder Tab..................................................................................................................................................................................6 Fig. 2 Plant Mask Tool Tab Before Color Segmentation.................................................................................................................................8 Fig. 3 Color picker tool...................................................................................................................................................................................8 Fig. 4 RGB Formula Computation tab page in Mask Detection Tools ............................................................................................................9 Fig. 5 Graph display performance of the selected threshold line ................................................................................................................10 Fig. 6 Machine Learning Computation tab page in Mask Detection Tools ..................................................................................................10 Fig. 7 Preprocessing Options for machine learning plant mask computation .............................................................................................11 Fig. 8 Plant mask settings for RGB Formula Segmentation Fig. 9 Plant mask settings for Machine Learning Computation ...................15 Fig. 10 Masking ............................................................................................................................................................................................15 Fig. 11 Plant Mask Tool Tab After Color Segmentation ...............................................................................................................................16 Fig. 12 Settings Tab......................................................................................................................................................................................17 Fig. 13 Color generating in progress (1) and finished (2) .............................................................................................................................18 Fig. 14 Single Run Tab..................................................................................................................................................................................19 Fig. 15 Batch Run Tab ..................................................................................................................................................................................19
Page | 21