Blue-and-white cauliflower ball color detection method based on high-throughput phenotype technology

Abstract

The invention provides a broccoli ball color detection method based on a high-throughput phenotype technology, and the method specifically comprises the steps: integrating a photosensitive sensor in an imaging unit of a field high-throughput phenotype platform, and remotely triggering an RGB imaging unit to carry out the image data collection of broccoli when the photosensitive sensor detects that the illumination intensity is in a preset range; the collected images are preprocessed; a pre-trained YOLOv8 target detection model is adopted to carry out ball-flower identification on the preprocessed image, a bounding box is generated according to an identified ball-flower area, and position coordinate information of the bounding box is recorded; classifying the colors of the ball-flower into different categories for the recognized ball-flower areas; and performing statistical analysis and visual output on the color data of the ball-flower, and storing all detection data. According to the method, the color data of the broccoli balls can be efficiently and accurately obtained, manual operation of breeding work is reduced, ambient light interference and personal errors are reduced, and the reliability and consistency of color detection are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of broccoli breeding, and specifically relates to a broccoli floret color detection method based on high-throughput phenotyping technology. BACKGROUND

[0002] Brassica oleracea var. italica (English name: Broccoli) is a Brassica oleracea plant, which is a kind of nutrient-rich and high-value vegetable crop and is widely planted worldwide. Broccoli floret color is one of its important phenotypic traits, which not only affects the appearance quality of commodities, but also is closely related to the content of antioxidant substances (such as carotenoids and chlorophyll), and is an important selection index in breeding work. However, the traditional floret color detection method mainly relies on manual observation or color card comparison, which has strong subjectivity, low efficiency, and is easily affected by environmental light, and cannot meet the needs of large-scale breeding and phenomics research.

[0003] In order to cultivate broccoli varieties with excellent floret color, breeders need to accurately measure and evaluate the floret color of different varieties or lines. Traditional methods usually require manual sampling, observation and recording, which not only consumes time and effort, but also easily leads to inconsistency of detection results due to operator fatigue or changes in environmental light. In addition, broccoli floret color is greatly affected by environmental factors such as light and temperature, and color detection under field conditions often cannot achieve standardization and scaling.

[0004] High-throughput phenotyping technology is a technology method that quickly obtains large amounts of plant phenotypic information through machine vision, sensor detection and intelligent algorithms, and has important application value in modern agricultural breeding and phenomics research. This technology can efficiently and non-destructively detect plant morphological, color, spectral characteristics and other phenotypic traits, and quantify color and other phenotypic parameters through analysis of multispectral or hyperspectral images. However, in the field of broccoli breeding and phenotyping research, traditional manual detection is still the main method, and the floret color detection method based on high-throughput phenotyping technology has not been fully developed and applied, and the related intelligent algorithms and standardized processes need further research. In view of this, the present application proposes a broccoli floret color detection method based on high-throughput phenotyping technology. SUMMARY

[0005] The purpose of the present application is to propose a broccoli floret color detection method based on high-throughput phenotyping technology, which adopts RGB imaging technology, machine vision recognition and artificial intelligence algorithms, and automatically detects the color characteristics of broccoli florets through field image acquisition and intelligent analysis, and quantifies the phenotypic information. This method can efficiently and accurately obtain the color data of broccoli florets, help reduce manual operations in breeding work, reduce environmental light interference and human error, and improve the reliability and consistency of color detection.

[0006] To achieve the above object, the technical scheme of the present application is: A broccoli floret color detection method based on high-throughput phenotyping technology, comprising the following steps: S1. Integrating a photosensitive sensor in an imaging unit of a field high-throughput phenotyping platform, and remotely triggering an RGB imaging unit to collect image data of broccoli when the photosensitive sensor detects that the light intensity is within a preset range; S2. Preprocessing the collected image; S3. Using a pre-trained YOLOv8 target detection model to identify the floret in the preprocessed image, generating a bounding box according to the identified floret region and recording the bounding box position coordinate information; and classifying the floret color into different categories for the identified floret region; S4. Statistically analyzing and visually outputting the floret color data, and storing all detection data.

[0007] Preferably, when the photosensitive sensor detects that the light intensity is within the range of 120-150 μmol / (m 2 ·s), the RGB imaging unit is remotely triggered to collect image data of the broccoli.

[0008] Preferably, the RGB imaging unit collects top view images of the broccoli plants and florets, and at least one complete broccoli plant is collected in a single image.

[0009] Preferably, the photosensitive sensor uses an Aqara light sensor.

[0010] Preferably, the preprocessing of the collected image includes denoising, background segmentation, and color correction.

[0011] Preferably, the preprocessing specifically includes noise elimination based on Gaussian filtering, background segmentation technology based on HSV color space, and color correction based on a standard color card.

[0012] Preferably, the construction of the detection data set for pre-training of the YOLOv8 target detection model is as follows: the collected broccoli images are made into xml format data sets; python is selected as the assembly language, LabelImg software is used to label the broccoli florets in the images with a rectangular bounding box, and the broccoli florets are color classified and labeled.

[0013] Preferably, the pre-trained YOLOv8 target detection model classifies the floret color into different categories, and the categories of the floret color include yellow-green florets, light green florets, green florets, gray-green florets, and blue-green florets.

[0014] Preferably, the statistical analysis and visual output of the flower ball color data specifically refers to: statistical analysis of the flower ball color data combined with image resolution and image size, and output of the analysis results in a visual form, including generation of a flower ball color histogram.

[0015] Preferably, the storage of all detection data specifically refers to: storage of the flower ball color detection data to a database, including original image data, preprocessing results, detection frame coordinates, color classification results and various statistical indicators; the database adopts a relational architecture design, supports fast query and batch export.

[0016] Compared with the prior art, the present application has the following beneficial effects: The present application adopts a high-throughput phenotype technology combined with RGB imaging and a YOLOv8 target detection algorithm, realizes intelligent and batch detection of the flower ball color of green flower, reduces manual operation in traditional color detection work, improves the accuracy and reliability of color data, and provides a high-throughput phenotype information acquisition method for the flower ball color of green flower. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 It is a visual schematic diagram of the field high-throughput phenotype platform of the present application. DETAILED DESCRIPTION

[0018] The technical solutions of the present application will be specifically described below. Figure 1 The technical solutions of the present application will be specifically described below.

[0019] The present application proposes a green flower flower ball color detection method based on high-throughput phenotype technology, including the following steps: S1, integrating a photosensitive sensor in the imaging unit of the field high-throughput phenotype platform, remotely triggering the RGB imaging unit to collect image data of mature green flower when the photosensitive sensor detects that the light intensity is in a preset range; the collected high-definition image (the default resolution is 640*640 pixels) is transmitted to the central computer in real time through a wireless network, and a preprocessing process is automatically started; S2, preprocessing the collected image; including denoising, background segmentation and color correction; S3, using a pre-trained YOLOv8 target detection model to recognize the flower ball of the preprocessed image, accurately positioning the green flower ball region, generating a boundary box according to the recognized flower ball region and recording the boundary box position coordinate information; for the recognized flower ball region, the flower ball color is classified into different categories; S4, statistical analysis and visual output of the flower ball color data, and storage of all detection data.

[0020] The present application trains the current classic detection network (including SSD, Faster R-CNN and YOLOv8) through model training, and evaluates and analyzes the trained broccoli floret detection model, and finally selects the YOLOv8 target detection model as the optimal detection network, so the YOLOv8 target detection algorithm is used in the present application.

[0021] The establishment of the detection data set is specifically: The collected broccoli image is made into an xml format data set. Python is selected as the assembly language, and LabelImg software is used to mark the broccoli florets in the image with a rectangular bounding box and color classification marking. At the same time, the data set is divided into training set, validation set and test set according to the ratio of 8:1:1.

[0022] Specifically, the training environment parameters are specifically: the picture resolution is uniformly set to 640x640, the batch size (Batchsize) is set to 4, the workers are 8, and YOLOv8s.pt is used as the training weight. In the process of model training, the initial learning rate (Learning Rate, LR) is set to 0.01, the Adam optimizer is used for dynamic adjustment, and the iteration number (Epoch) of the model in the training process is 200. When performing performance testing, the IoU threshold is set to 0.5, and the confidence threshold conf is set to 0.5.

[0023] In the embodiment, when the photosensitive sensor detects that the light intensity is in the range of 120~150μmol / (m 2 ·s), the RGB imaging unit is remotely triggered to collect image data of broccoli.

[0024] In the embodiment, the RGB imaging unit collects top view images of broccoli plants and florets, and at least one complete broccoli plant is collected in a single image.

[0025] In the embodiment, the photosensitive sensor uses Aqara light sensor, which can be connected with mobile phone for real-time monitoring of environmental spectrum changes. The light intensity detected by the photosensitive sensor is in the range of 120~150μmol / (m2·s), and in this range, the color of the color card of the broccoli floret is consistent with the image color shot by the high-throughput phenotype platform.

[0026] In the embodiment, advanced image processing algorithms are used in the preprocessing stage, including noise elimination based on Gaussian filtering, background segmentation technology based on HSV color space, and color correction based on standard color card.

[0027] In this embodiment, the color classification stage adopts a fine-grained classification method based on deep learning. For the identified flower ball region, a pre-trained YOLOv8 target detection model is used for flower ball color classification. This model can accurately classify flower ball colors into five professional categories: yellow-green (YG), light green (LG), standard green (G), gray-green (GG), and blue-green (BG). A strict confidence threshold (0.5) and IoU threshold (0.5) are set during the classification process to ensure the reliability of the classification results. The system also automatically records the confidence scores of each classification result to provide a reference for subsequent analysis. Through this classification method, flower ball colors can be directly recognized and classified during the detection process.

[0028] In this embodiment, the statistical analysis and visual output of the flower ball color data are as follows: combining image resolution and image size, the position information of each flower ball is calculated, the flower ball color data is statistically analyzed, and the analysis results are output in a visual form, including generating a flower ball color histogram. These charts can visually display the distribution and proportion of different color categories in the field, providing important decision-making basis for researchers.

[0029] In this embodiment, the storage of all detection data is as follows: flower ball color detection data, including original image data, preprocessing results, detection box coordinates, color classification results, and various statistical indicators, are stored in a database. The database uses a relational architecture design, supporting fast query and batch export. It supports subsequent variety comparison, genetic analysis, and breeding selection. Researchers can view the detection results in real time through a dedicated client software, or generate various analysis reports as needed. The system also provides data comparison functions, which can easily compare and analyze different varieties and different growth stages, providing strong data support for breeding work. Through real-time computer viewing of detection results, intelligent and real-time flower ball color detection is achieved.

[0030] A specific embodiment is provided below: The overall structure of the field high-throughput phenotype platform shown in FIG. 1, in which an Aqara photosensitive sensor is integrated in the imaging unit. When the Aqara light sensor detects that the ambient light intensity reaches 135 μmol / (m²·s), the system automatically triggers the image acquisition device to obtain high-definition images. After the collected image data is preprocessed by the platform algorithm, the required data can be copied to the local computer through the external mobile hard disk. Subsequently, based on the established labeled data set, the YOLOv8 target detection algorithm is used for intelligent analysis, successfully identifying 3 flower ball regions, and generating accurate boundary box coordinates (x1=125, y1=86, x2=203, y2=165; x1=312, y1=94, x2=389, y2=178; x1=498, y1=102, x2=576, y2=190). Color classification is performed on the first flower ball region, and the model output confidence is 0.92, which is determined as the light green (LG) category; the second flower ball region confidence is 0.88, which is determined as the green (G) category; the third flower ball region confidence is 0.85, which is determined as the blue-green (BG) category. The system automatically counts the detection results this time: the LG category accounts for 23.4%, the G category accounts for 33.3%, and the BG category accounts for 43.3%, and generates the corresponding color distribution histogram. All detection data including the original image, boundary box coordinates, color classification results and statistical indicators are automatically stored in the database, and the analysis report can be viewed and exported in real time through the client software.

[0031] In summary, the present application discloses a broccoli floret color detection method based on high-throughput phenotype technology, aiming to realize rapid, accurate and non-destructive detection of broccoli floret color through high-resolution imaging technology and computer vision algorithm. This method integrates a photosensitive sensor in the imaging unit of the field high-throughput phenotype platform and connects it with a mobile phone, realizing intelligent and real-time floret color detection. The photosensitive sensor can monitor the changes in the environmental spectrum in real time, and when a specific spectral range is detected, the mobile phone will send a prompt, and the user will manually trigger high-resolution image acquisition. The collected data will be automatically stored in a standardized format on the computer, and the user can conveniently copy the required data through an external mobile hard disk. This method effectively solves the color distortion problem caused by changes in environmental light in traditional detection methods, while improving the convenience and efficiency of data acquisition. Through computer vision algorithms, the collected images are analyzed for color feature extraction and analysis, achieving accurate and non-destructive detection of broccoli floret color. The present application provides efficient and intelligent technical support for broccoli quality evaluation, genetic breeding and phenotypeomics research, and has wide application prospects.

[0032] The above is the preferred embodiment of the present application, any changes made in accordance with the technical solutions of the present application, as long as the resulting function does not exceed the scope of the technical solutions of the present application, belongs to the protection scope of the present application.

Claims

1. A method for detecting broccoli floret color based on high-throughput phenotyping technology, characterized in that, The method comprises the following steps: S1. Integrating a photosensitive sensor in an imaging unit of a field high-throughput phenotyping platform, and remotely triggering an RGB imaging unit to collect image data of broccoli when the photosensitive sensor detects that the light intensity is in a preset range; S2. Preprocessing the collected image; S3. Identifying the flower ball in the preprocessed image by using a pre-trained YOLOv8 target detection model, generating a bounding box according to the identified flower ball region and recording the position coordinate information of the bounding box; and classifying the flower ball color into different categories for the identified flower ball region; S4. Statistically analyzing and visually outputting the flower ball color data, and storing all detection data.

2. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, When the photosensitive sensor detects that the light intensity is in the range of 120~150 μmol / (m 2 ·s), the remote trigger RGB imaging unit performs image data acquisition on the broccoli.

3. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, The RGB imaging unit collects top-view images of broccoli plants and flower balls, and at least one complete broccoli plant is collected in a single image.

4. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, The photosensitive sensor adopts an Aqara light sensor.

5. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, The preprocessing of the collected image comprises denoising, background segmentation and color correction.

6. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 5, characterized in that, The preprocessing specifically comprises noise elimination based on Gaussian filtering, background segmentation technology based on HSV color space, and color correction based on a standard color card.

7. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, The construction of the detection data set for pre-training of the YOLOv8 target detection model specifically comprises: preparing the collected broccoli images into an xml format data set; selecting python as the assembly language, using LabelImg software to mark the broccoli flower balls in the images in the form of a rectangular bounding box, and performing color classification marking on the broccoli flower balls.

8. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, In the classification of the flower ball color by using the pre-trained YOLOv8 target detection model, the categories of the flower ball color include yellow-green flower balls, light green flower balls, green flower balls, gray-green flower balls and blue-green flower balls.

9. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, The statistical analysis and visual output of the flower ball color data specifically comprise: statistically analyzing the flower ball color data in combination with the image resolution and the image size, and outputting the analysis results in a visual form, including generating a flower ball color histogram.

10. The broccoli floret color detection method based on high-throughput phenotyping technology according to claim 1, characterized in that, Storing all detection data specifically comprises: storing the flower ball color detection data into a database, including original image data, preprocessing results, detection box coordinates, color classification results and various statistical indicators; The database adopts a relational architecture design and supports fast query and batch export.

Images