A Chinese cabbage phenotype detection system and method

Through the image and odor acquisition module of the Chinese cabbage phenotype detection system, combined with the YOLOv5s network and electronic nose, the problems of low manual detection efficiency and poor accuracy are solved, and the automated detection of cabbage phenotype and internal rot are realized, which improves the detection efficiency and Accuracy, facilitates healthy planting of cabbage.

CN117853469BActive Publication Date: 2024-12-03HEBEI AGRICULTURAL UNIV.
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410068036.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-17
Publication Date
2024-12-03
Estimated Expiration
2044-01-17

AI Technical Summary

Technical Problem

In the prior art, cabbage phenotype detection relies on manual inspection, which is time-consuming and labor-intensive and has low accuracy, making it difficult to identify the internal rot of cabbage, resulting in poor results in preventing and treating diseases.

Method used

Design a phenotype detection system for cabbage, including image acquisition module, phenotype detection module, odor collection module and odor recognition module. By building a phenotype detection model based on YOLOv5s network and electronic nose, it realizes the collection of cabbage appearance images and odor information. and analyze to determine whether there is abnormality and internal rot of cabbage.

Benefits of technology

The phenotype detection and internal rot detection of cabbage has been automated, the detection efficiency and accuracy have been improved, the abnormal situation of cabbage can be adjusted in a timely manner, and the healthy cultivation and cultivation of cabbage can be promoted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117853469B_ABST
    Figure CN117853469B_ABST
Patent Text Reader

Abstract

The present invention provides a Chinese cabbage phenotype detection system and method, the method comprising: collecting the appearance image of the cabbage through an image acquisition module, collecting the odor information of the cabbage through an odor acquisition module, building a phenotype detection model through a phenotype detection module, establishing a phenotype detection training data set, training the phenotype detection model based on the phenotype detection training data set to obtain a trained phenotype detection model, inputting the collected appearance image of the cabbage into the trained phenotype detection model to obtain a phenotype detection result, and obtaining the odor information of the collected cabbage through an odor recognition module, and comparing it with a preset standard odor of the cabbage to determine whether the inside of the cabbage is rotten. The Chinese cabbage phenotype detection system and method provided by the present invention can realize the phenotype detection of the cabbage and realize the internal rot detection based on the odor, can timely adjust the cabbage according to the abnormality, is convenient for the cultivation and planting of the cabbage, and is convenient for use.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of cabbage phenotype detection, and in particular to a Chinese cabbage phenotype detection system and method. Background Art

[0002] During the planting process, cabbage may develop diseases due to various reasons, which may lead to reduced yield and quality. Therefore, it is often necessary to perform phenotypic detection on it to achieve scientific disease prevention and control. However, in the prior art, manual inspection is often required, and manual phenotypic detection is performed based on the experience of growers, which is not only time-consuming and laborious, but also has a low accuracy rate and poor effect. In addition, it is also difficult to identify the rot inside the cabbage. Therefore, it is very necessary to design a cabbage phenotypic detection system and method. Summary of the invention

[0003] The purpose of the present invention is to provide a Chinese cabbage phenotype detection system and method, which can realize the phenotypic detection of cabbage and internal decay detection based on odor, and can timely adjust the cabbage according to its abnormality, so as to facilitate the cultivation and planting of cabbage and facilitate use.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A Chinese cabbage phenotype detection system comprises: an image acquisition module, a phenotype detection module, an odor acquisition module, an odor recognition module and a control module, wherein the image acquisition module, the phenotype detection module, the odor acquisition module and the odor recognition module are connected to the control module;

[0006] The image acquisition module is used to acquire the appearance image of the cabbage;

[0007] The phenotype detection module is used to judge whether the cabbage has abnormalities based on the collected appearance images of the cabbage;

[0008] The odor collection module is used to collect odor information of cabbage;

[0009] The odor recognition module is used to determine whether the inside of the cabbage is rotten according to the collected odor information of the cabbage.

[0010] Optionally, the odor collection module is an electronic nose.

[0011] The present invention also provides a Chinese cabbage phenotype detection method, which is applied to the above-mentioned Chinese cabbage phenotype detection system and comprises the following steps:

[0012] Step 1: The appearance image of the cabbage is collected by the image collection module, and the odor information of the cabbage is collected by the odor collection module;

[0013] Step 2: The phenotypic detection module builds a phenotypic detection model;

[0014] Step 3: The phenotype detection module establishes a phenotype detection training data set;

[0015] Step 4: training the phenotype detection model based on the phenotype detection training data set to obtain a trained phenotype detection model, inputting the collected appearance image of the Chinese cabbage into the trained phenotype detection model to obtain a phenotype detection result;

[0016] Step 5: The odor recognition module obtains the odor information of the collected cabbage and compares it with the preset standard odor of cabbage to determine whether the inside of the cabbage is rotten.

[0017] Optionally, in step 2, the phenotype detection module builds a phenotype detection model, specifically:

[0018] A phenotypic detection model is built based on the improved YOLOv5s network. The phenotypic detection model includes a backbone network, a bottleneck network, an InceptionA module and a detection head. The InceptionA module is added between the backbone network and the bottleneck network. The backbone network is a CSPDarknet network, the bottleneck network is a fusion structure of FPN and PANet, and the detection head is an upsampling and splicing structure.

[0019] Optionally, in step 3, the phenotype detection module establishes a phenotype detection training data set, specifically:

[0020] The images of cabbage are collected, and the images of cabbage are annotated by using an annotation tool. After all images are annotated, the annotated images are preprocessed, wherein the preprocessing includes geometric transformation and saturation and brightness adjustment.

[0021] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the Chinese cabbage phenotype detection system and method provided by the present invention, the system comprises an image acquisition module, a phenotype detection module, an odor acquisition module, an odor recognition module and a control module, the method comprises collecting the appearance image of the cabbage through the image acquisition module, collecting the odor information of the cabbage through the odor acquisition module, building a phenotype detection model by the phenotype detection module, establishing a phenotype detection training data set by the phenotype detection module, training the phenotype detection model based on the phenotype detection training data set to obtain a trained phenotype detection model, inputting the collected appearance image of the cabbage into the trained phenotype detection model to obtain a phenotype detection result, the odor recognition module obtains the collected odor information of the cabbage, and compares it with a preset standard odor of the cabbage to determine whether the inside of the cabbage is rotten; the system and method can realize the phenotypic detection of the cabbage and realize the internal rot detection based on the odor, and can timely adjust the cabbage according to the abnormality, which is convenient for the cultivation and planting of the cabbage and easy to use. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of the structure of a Chinese cabbage phenotype detection system provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a method for detecting phenotype of Chinese cabbage provided in an embodiment of the present application;

[0024] Figure 3 It is a schematic diagram of the structure of the phenotypic detection model;

[0025] Figure 4 This is a schematic diagram of the InceptionA module structure. DETAILED DESCRIPTION

[0026] The purpose of the present invention is to provide a Chinese cabbage phenotype detection system and method, which can realize the phenotypic detection of cabbage and internal decay detection based on odor, and can timely adjust the cabbage according to its abnormality, so as to facilitate the cultivation and planting of cabbage and facilitate use.

[0027] like Figure 1 As shown, the Chinese cabbage phenotype detection system provided by the embodiment of the present invention includes: an image acquisition module, a phenotype detection module, an odor acquisition module, an odor recognition module and a control module, wherein the image acquisition module, the phenotype detection module, the odor acquisition module and the odor recognition module are connected to the control module;

[0028] The image acquisition module is used to acquire the appearance image of the cabbage;

[0029] The phenotype detection module is used to judge whether the cabbage has abnormalities based on the collected appearance images of the cabbage;

[0030] The odor collection module is used to collect odor information of cabbage;

[0031] The odor recognition module is used to determine whether the inside of the cabbage is rotten according to the collected odor information of the cabbage.

[0032] The odor collection module is an electronic nose.

[0033] The image acquisition module can adopt a camera or the like.

[0034] like Figure 2 As shown, the present invention also provides a Chinese cabbage phenotype detection method, which is applied to the above-mentioned Chinese cabbage phenotype detection system and comprises the following steps:

[0035] Step 1: The appearance image of the cabbage is collected by the image collection module, and the odor information of the cabbage is collected by the odor collection module;

[0036] Step 2: The phenotypic detection module builds a phenotypic detection model;

[0037] Step 3: The phenotype detection module establishes a phenotype detection training data set;

[0038] Step 4: training the phenotype detection model based on the phenotype detection training data set to obtain a trained phenotype detection model, inputting the collected appearance image of the Chinese cabbage into the trained phenotype detection model to obtain a phenotype detection result;

[0039] Step 5: The odor recognition module obtains the odor information of the collected cabbage and compares it with the preset standard odor of cabbage to determine whether the inside of the cabbage is rotten.

[0040] In step 2, the phenotype detection module builds a phenotype detection model, specifically:

[0041] A phenotype detection model is built based on the improved YOLOv5s network. The phenotype detection model includes a backbone network, a bottleneck network, an InceptionA module and a detection head. The InceptionA module is added between the backbone network and the bottleneck network. The backbone network is a CSPDarknet network, the bottleneck network is a fusion structure of FPN and PANet, and the detection head is an upsampling and splicing structure.

[0042] The phenotype detection model is introduced in detail. First, the network structure diagram of the phenotype detection model is as follows Figure 3 As shown in the figure, the CSPDarknet network is used to extract rich features from the input image. The CSPDarknet network splits the feature map into two parts, one of which is directly convolved, and the other is spliced ​​with the result of the previous convolution. Compared with other target detection networks in the prior art, it can integrate gradient changes into the feature map, eliminate repeated network gradient information, and reduce model parameters and the number of floating-point operations; the structure of the CSPDarknet network adopts two CSP structures, namely CSP1_X module and CSP2_X module, wherein the CSP1_X module is spliced ​​by CBS, Resunit and convolution layer, and is used in the backbone network, and the CSP2_X module is spliced ​​by convolution layer and multiple Resunit, and is used in the bottleneck network, wherein Resunit is a residual structure based on ResNet;

[0043] The bottleneck network adopts the FPN and PANet fusion structure, and strengthens the fusion ability of network features through the CSP2_X module. FPN is used to transfer the semantic features with strong deep layers to the shallow layers, enhancing the expression of features at multiple scales, and PANet is used to transfer the location information with strong shallow layers to the deep layers, enhancing the positioning ability of the network.

[0044] The InceptionA module used in the present invention is as follows Figure 4As shown in the figure, the InceptionA module is added between the backbone network and the bottleneck network, and connected to the second C3, the third C3 and the SPP module, so that the feature map is first input into the InceptionA module before fusion. After extracting multi-scale features, a small convolution is used to obtain a small receptive field. The network can more accurately perceive the small-scale disease features. The features of the four scales are fused and then spliced ​​and output, so that the network can still pay attention to large-scale diseases and obtain richer feature information. The feature map size of the second C3 module is [32, 32, 256 ], connected to the InceptionA module using 4 branches for feature extraction, the channel number ratio is 2:2:3:1, the feature map size of the third C3 module is [32, 32, 512], and the feature map size of the SPP module is [32, 32, 1024]. Compared with the shallow feature map, the deep feature map contains more fine-grained features. The InceptionA module can retain more feature information and improve the network's detection effect on small-scale diseases and densely diseased plants. In addition, the InceptionA module can also improve the detection accuracy of the network.

[0045] For the loss function of the phenotype detection model, the commonly used loss functions include IoU, DIoU and GIoU. The mathematical expression of the IoU loss function is:

[0046]

[0047] In the formula, A g Represents the area of ​​the real box, A p Indicates the area of ​​the predicted box. IoU may have the following two problems: when the two candidate boxes have no intersection with the real box, the calculated IoU_Loss is 0 and the loss is 1, but there must be one candidate box that is closer to the real box than the other candidate box. When the IoU of the two candidate boxes and the real box is the same, but the positions of the two candidate boxes are different, it is impossible to determine which candidate box is more accurate;

[0048] In order to solve the first problem in IoU, the GIoU loss function introduces the minimum closure area, that is, the minimum enclosing rectangle of the real box and the predicted box and the difference between the enclosing rectangle and the union, which improves the way to measure the degree of intersection between the candidate box and the real box. However, when the height and width of the two candidate boxes are the same and they are in the same plane, GIoU degenerates into IoU. At the same time, both GIoU and IoU have the problems of slow convergence and inaccurate regression. The mathematical expression of the GIoU loss function is:

[0049]

[0050] In the formula, A g Represents the area of ​​the real box, Ap Represents the area of ​​the prediction box, A c represents the area of ​​the smallest rectangle containing the real box and the predicted box, and u represents the area of ​​the intersection of the real box and the predicted box;

[0051] In response to the problems of IoU and GIoU, the DIoU loss function solves the problem of the distance between the minimum normalized candidate box and the true box. When the candidate box does not overlap with the true box, DIoU can indicate the direction of movement of the bounding box. When the true box contains the candidate box, DIoU will directly measure the distance between the two candidate boxes. Compared with GIoU, it converges faster. For two candidate boxes in the horizontal and vertical directions, DIoU will consider the overlapping area between the candidate box and the true box and the distance to the center point, so the regression speed is also very fast. However, when the center point of the candidate box coincides with the center point of the true box, the intersection and union ratio is the same, and the aspect ratio is different, the calculated DIoU, GIoU and IoU are the same. The mathematical expression of the DIoU loss function is:

[0052]

[0053] In the formula, A g Represents the area of ​​the real box, A p represents the area of ​​the prediction box, ρ(·) represents the Euclidean distance, and b g represents the center point of the real box, b p represents the center point of the prediction box, and c represents the diagonal length of the minimum rectangle containing the real box and the prediction box;

[0054] The CIoU loss function comprehensively considers the overlapping area, center point distance, and aspect ratio of the candidate box, so that the size of the candidate box is more consistent with the real box. The mathematical expression of the CIoU loss function is:

[0055]

[0056]

[0057]

[0058] In the formula, A g Represents the area of ​​the real box, A p represents the area of ​​the prediction box, ρ(·) represents the Euclidean distance, and b g represents the center point of the real box, b p represents the center point of the predicted box, c represents the diagonal length of the smallest rectangle containing the real box and the predicted box, α×v is the penalty term for the aspect ratio, α is a positive number, v is used to measure the similarity of the aspect ratio, ω g Indicates the width of the real box, h g Represents the height of the real box, ω pIndicates the width of the prediction box, h p Indicates the height of the prediction box;

[0059] IoU mainly considers the overlapping area between the candidate box and the real box. GIoU solves the problem when the bounding boxes do not overlap based on IOU. DIoU also considers the distance between the bounding box and the center point based on IOU and GIoU. By integrating the advantages of the first three loss functions, CIoU comprehensively considers the overlapping area, center point distance and aspect ratio of the candidate box. In order to achieve the best detection effect, CIoU is used as the loss function of YOLOv5sIA.

[0060] In step 3, the phenotype detection module establishes a phenotype detection training data set, specifically:

[0061] Collect images of cabbage, and use the labellmg annotation tool to annotate the images of cabbage. After all images are annotated, preprocess the annotated images. The preprocessing includes geometric transformation and saturation and brightness adjustment. The set transformation includes rotation, horizontal mirroring and vertical mirroring, which are described in detail as follows:

[0062] Image rotation is to rotate the image by a certain angle to obtain a new image. The points before and after rotation remain unchanged from the center.

[0063] Horizontal mirroring is to exchange all the pixel values ​​of the left and right parts with the central vertical bisector of the image as the symmetry axis to obtain a new image.

[0064] Vertical mirroring is to exchange all the pixel values ​​of the upper and lower parts of the image with the central horizontal bisector of the image as the symmetry axis to obtain a new image.

[0065] Saturation refers to the vividness of the color in the image, that is, the purity of the color. Its value is equal to chroma divided by lightness. The value depends on the ratio of the color component to the achromatic component (gray). The larger the color component, the greater the saturation; conversely, the smaller it is. Brightness refers to the brightness of the image. If its value is too large, people will not be able to see the details of the dark part of the image clearly.

[0066] Adjusting the image brightness is achieved by randomly increasing or decreasing the pixel RGB value of a certain point in the image.

[0067] Adjusting the saturation of an image requires converting the image from RGB to HSL, changing the size of the image's saturation value, and then converting it back from HSL to RGB.

[0068] Finally, data enhancement is performed based on Mosaic. Four images are randomly read from the training set each time, and these four images are scaled and horizontally flipped respectively. Then the images are arranged in the order of acquisition at the upper left, lower left, lower right, and upper right positions. After the four images are placed, the fixed areas of the four images are intercepted using a matrix method, and finally they are spliced ​​into one complete image. The spliced ​​image contains four real frames, thereby enriching the sample size of the training set.

[0069] In step 5, the odor recognition module obtains the odor information of the collected cabbage and compares it with the preset standard odor of cabbage to determine whether the inside of the cabbage is rotten, specifically:

[0070] The electronic nose collects the smell information of the cabbage and sends it to the smell recognition module through the control module. The smell recognition module can directly compare the collected smell information with the standard smell of the cabbage, judge whether the cabbage is rotten according to the comparison result, and judge whether the inside of the cabbage is rotten in combination with the phenotypic detection result. If the phenotypic detection result is abnormal, no judgment is required. If the phenotypic result is normal and the comparison result is normal, the cabbage is judged to be normal. If the phenotypic result is normal but the comparison result is abnormal, it is judged that the inside of the cabbage is rotten.

[0071] In this regard, an odor recognition model can also be built based on a deep learning network to collect relevant odor information of cabbage and annotate it to generate an odor recognition data set. The odor recognition model can be trained based on the odor recognition data set to obtain a trained odor recognition model. The collected odor information of cabbage can be input into the trained odor recognition model to obtain an odor recognition result.

[0072] The present invention provides a Chinese cabbage phenotype detection system and method. The system comprises an image acquisition module, a phenotype detection module, an odor acquisition module, an odor recognition module and a control module. The method comprises the following steps: acquiring an appearance image of the Chinese cabbage through the image acquisition module, acquiring odor information of the Chinese cabbage through the odor acquisition module, building a phenotype detection model through the phenotype detection module, establishing a phenotype detection training data set, training the phenotype detection model based on the phenotype detection training data set to obtain a trained phenotype detection model, inputting the acquired appearance image of the Chinese cabbage into the trained phenotype detection model to obtain a phenotype detection result, and acquiring the odor information of the acquired Chinese cabbage through the odor recognition module, and comparing the odor information with a preset standard odor of the Chinese cabbage to determine whether the inside of the Chinese cabbage is rotten. The system and method can realize phenotypic detection of the Chinese cabbage and realize internal rot detection based on the odor, can timely adjust the Chinese cabbage according to its abnormality, and is convenient for the cultivation and planting of the Chinese cabbage and for use.

Claims

1. A Chinese cabbage phenotype detection system, characterized in that: include: An image acquisition module, a phenotype detection module, an odor acquisition module, an odor recognition module, and a control module, wherein the image acquisition module, the phenotype detection module, the odor acquisition module, and the odor recognition module are connected to the control module; The image acquisition module is used to acquire the appearance image of the cabbage; The phenotype detection module is used to judge whether the cabbage has abnormalities based on the collected appearance images of the cabbage; The odor collection module is used to collect odor information of cabbage; The odor recognition module is used to determine whether the inside of the cabbage is rotten according to the collected odor information of the cabbage; The phenotypic detection module builds a phenotypic detection model, specifically: A phenotypic detection model is built based on the improved YOLOv5s network. The phenotypic detection model includes a backbone network, a bottleneck network, an InceptionA module, and a detection head. The backbone network is a CSPDarknet network, the bottleneck network is a fusion structure of FPN and PANet, and the detection head is an upsampling and splicing structure. The CSPDarknet network is used to extract rich features from the input image. The CSPDarknet network splits the feature map into two parts, one of which is directly convolved, and the other is spliced ​​with the result of the previous convolution. The structure of the CSPDarknet network adopts two CSP structures, namely CSP1_X module and CSP2_X module. Among them, the CSP1_X module is composed of CBS, Resunit and convolution layer, and is used in the backbone network. The CSP2_X module is composed of convolution layer and multiple Resunit, and is used in the bottleneck network. Among them, Resunit is a residual structure based on ResNet. The bottleneck network adopts the FPN and PANet fusion structure, and strengthens the fusion ability of network features through the CSP2_X module. FPN is used to transfer the semantic features with strong deep layers to the shallow layers, enhancing the expression of features at multiple scales, and PANet is used to transfer the location information with strong shallow layers to the deep layers, enhancing the positioning ability of the network. The InceptionA module is added between the backbone network and the bottleneck network, and connected to the second C3, the third C3 and the SPP modules, so that the feature map is first input into the InceptionA module before fusion. After extracting multi-scale features, a small convolution is used to obtain a small receptive field. The network can more accurately perceive the small-scale disease characteristics. The features of the four scales are fused and then spliced ​​and output, so that the network can still pay attention to large-scale diseases and obtain richer feature information. The feature map size of the second C3 module is [32, 32, 256], which is connected to the InceptionA module using 4 branches for feature extraction, and the channel number ratio is 2:2:3:

1. The feature map size of the third C3 module is [32, 32, 512], and the feature map size of the SPP module is [32, 32, 1024]. Compared with the shallow feature map, the deep feature map contains more fine-grained features.

2. The Chinese cabbage phenotype detection system according to claim 1, characterized in that: The odor collection module is an electronic nose.

3. A method for detecting Chinese cabbage phenotype, applied to the Chinese cabbage phenotype detection system according to any one of claims 1-2, characterized in that: The steps include: Step 1: The appearance image of the cabbage is collected by the image collection module, and the odor information of the cabbage is collected by the odor collection module; Step 2: Phenotype detection module builds a phenotype detection model, specifically: A phenotype detection model is built based on the improved YOLOv5s network. The phenotype detection model includes a backbone network, a bottleneck network, an InceptionA module and a detection head. The InceptionA module is added between the backbone network and the bottleneck network. The backbone network is a CSPDarknet network, the bottleneck network is a fusion structure of FPN and PANet, and the detection head is an upsampling and splicing structure. The CSPDarknet network is used to extract rich features from the input image. The CSPDarknet network splits the feature map into two parts, one of which is directly convolved, and the other is spliced ​​with the result of the previous convolution. The structure of the CSPDarknet network adopts two CSP structures, namely CSP1_X module and CSP2_X module. Among them, the CSP1_X module is composed of CBS, Resunit and convolution layer, and is used in the backbone network. The CSP2_X module is composed of convolution layer and multiple Resunit, and is used in the bottleneck network. Among them, Resunit is a residual structure based on ResNet. The bottleneck network adopts the FPN and PANet fusion structure, and strengthens the fusion ability of network features through the CSP2_X module. FPN is used to transfer the semantic features with strong deep layers to the shallow layers, enhancing the expression of features at multiple scales, and PANet is used to transfer the location information with strong shallow layers to the deep layers, enhancing the positioning ability of the network. The InceptionA module is added between the backbone network and the bottleneck network, and connected to the second C3, third C3 and SPP modules, so that the feature map is first input into the InceptionA module before fusion. After extracting multi-scale features, a small convolution is used to obtain a small receptive field. The network can more accurately perceive the small-scale disease features. The features of the four scales are fused and then spliced ​​and output, so that the network can still pay attention to large-scale diseases and obtain richer feature information. The feature map size of the second C3 module is [32, 32, 256], which is connected to the InceptionA module using 4 branches for feature extraction, and the channel number ratio is 2:2:3:

1. The feature map size of the third C3 module is [32, 32, 512], and the feature map size of the SPP module is [32, 32, 1024]. Compared with the shallow feature map, the deep feature map contains more fine-grained features; Step 3: The phenotype detection module establishes a phenotype detection training data set; Step 4: training the phenotype detection model based on the phenotype detection training data set to obtain a trained phenotype detection model, inputting the collected appearance image of the Chinese cabbage into the trained phenotype detection model to obtain a phenotype detection result; Step 5: The odor recognition module obtains the odor information of the collected cabbage and compares it with the preset standard odor of cabbage to determine whether the inside of the cabbage is rotten.

4. The method for detecting Chinese cabbage phenotype according to claim 3, characterized in that: In step 3, the phenotype detection module establishes a phenotype detection training data set, specifically: The images of cabbage are collected, and the images of cabbage are annotated by using an annotation tool. After all images are annotated, the annotated images are preprocessed, wherein the preprocessing includes geometric transformation and saturation and brightness adjustment.