Method and system for identifying sex of protaetia brevitarsis based on computer vision

Abstract

The invention discloses a protaetia brevitarsis sex identification method and system based on computer vision, and relates to the field of insect sex identification, and the method comprises the steps: S1, obtaining a high-definition digital image of a protaetia brevitarsis adult at least comprising a back, an abdomen and a forefoot shin segment area; s2, processing the acquired digital image; s3, segmenting a forefoot shin segment region and a web region from the preprocessed image; s4, extracting morphological features from the segmented regions, wherein the morphological features at least comprise glossiness features, forefoot tibia tooth profile features and web morphological features; s5, inputting the extracted morphological features into a pre-trained classification model; according to the method, by comprehensively utilizing the multi-dimensional external morphological characteristics such as glossiness, front foot shin section tooth shape, web shape and stripes stably existing on the body surface of the adult protaetia brevitarsis, non-destructive living body identification of the protaetia brevitarsis is achieved, and the follow-up utilization value of the protaetia brevitarsis in breeding and scientific research is guaranteed.

Description

Technical Field

[0001] This invention relates to insect sex identification technology, specifically to a method and system for sex identification of the white-spotted flower beetle based on computer vision. Background Technology

[0002] The white-spotted flower beetle (Protaetia brevitarsis, synonym Potossia brevitarsis) belongs to the family Scarabaeidae, subfamily Cetoninae, order Coleoptera. This insect possesses dual attributes as both an "agricultural pest" and a "resource insect." Precise population control and resource utilization are key research areas in the field of green agricultural cycles. Sexing technology, as a crucial link between precise control and resource utilization, can simultaneously support the efficient advancement of both directions and serves as the foundation for subsequent related research and applications.

[0003] Currently, the lack of sex identification technology makes it impossible to carry out targeted "precision control." For example, it is difficult to accurately monitor the ratio of male to female populations in the field, predict the trend of population outbreaks, and formulate differentiated control strategies for the female insect's egg-laying period and the male insect's gathering and damage period, which easily leads to low control efficiency and excessive use of pesticides.

[0004] The decomposition of agricultural waste by the larvae of the white-spotted flower beetle is a core aspect of its resource utilization, and the key to large-scale larval farming lies in the breeding efficiency of the breeding stock. The male-to-female ratio of the white-spotted flower beetle breeding stock directly affects the mating success rate, egg production, and larval hatching rate. If the sex cannot be quickly identified, it is difficult to accurately control the male-to-female ratio of the breeding stock, which can easily lead to problems such as male-to-female imbalance, low egg production, and unstable larval yield, thus restricting the efficiency of larval waste conversion and the stable supply of raw materials for bio-organic fertilizer.

[0005] There are two main methods for sexing existing insects: one is a destructive method that relies on dissecting the genitals. While accurate, this method results in the death of adult insects and cannot be applied to scenarios requiring the preservation of live specimens for breeding, behavioral observation, or resource utilization, thus limiting its practicality. The other method relies on experience-based visual identification. However, this method is subjective, varies from person to person, has low efficiency and poor stability, and requires highly experienced operators, making it difficult to promote in large-scale breeding or large-area monitoring. This can lead to problems such as unreasonable population ratios and inaccurate monitoring data.

[0006] Therefore, there is an urgent need for a precise, efficient, and convenient method to distinguish the sex of the white-spotted flower beetle, in order to overcome the shortcomings of existing methods. Summary of the Invention

[0007] The purpose of this invention is to provide a computer vision-based method and system for sexing white-spotted beetles, in order to solve the problem that existing technologies rely on destructive dissection or subjective experience, which cannot simultaneously meet the requirements of non-destructive, rapid, accurate and standardized operation.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] In a first aspect, the present invention provides a computer vision-based method for sexing white-spotted scarab beetles, comprising:

[0010] S1. Obtain high-resolution digital images of adult white-spotted flower beetles, including at least the dorsal, abdominal, and fore tibia regions;

[0011] S2. Denoise, enhance, and standardize the acquired digital images;

[0012] S3. Segment the foreleg tibia region and ventral region from the preprocessed image;

[0013] S4. Extract morphological features from the segmented region, wherein the morphological features include at least gloss features, foreleg tibia tooth shape features, and sternite morphological features;

[0014] S5. Input the extracted morphological features into the pre-trained classification model and output the sex determination result.

[0015] In step S3, the segmentation of the foreleg tibia region is based on color space conversion and threshold segmentation, while the segmentation of the ventral region is based on edge detection and region growing.

[0016] Furthermore, in step S4, the gloss feature is obtained by calculating the statistical features of the color channel of the dorsal region of the insect body, and is used to classify the insect body into a glossy category or a matte category.

[0017] Furthermore, in step S4, the extraction of the forefoot tibia tooth shape features includes the following sub-steps:

[0018] Edge detection was performed on the forefoot tibia region image to extract contour features;

[0019] The outlines of the three teeth are selected, and the centroid coordinates of the outline of each tooth are calculated.

[0020] Calculate the distance between the centroids of the first and second teeth, and the distance between the centroids of the second and third teeth;

[0021] Calculate the ratio of the contour area of ​​the third tooth to that of the first tooth;

[0022] Calculate the roundness of the third tooth profile;

[0023] Distance, area ratio, and roundness were used as characteristics of the forefoot tibia teeth.

[0024] Furthermore, in step S4, the extraction of web morphological features includes the following sub-steps:

[0025] The image of the abdominal region is converted to grayscale.

[0026] Calculate the grayscale projection curve of the web region in the vertical direction;

[0027] The grayscale projection curve is smoothed and its first derivative is calculated to determine the web concave region.

[0028] The depth of the depression and the percentage of the depression area are calculated as morphological characteristics of the web.

[0029] Furthermore, in step S4, the extracted morphological features also include surface markings.

[0030] The surface markings were obtained through white marking detection and cluster analysis, including the number of markings and the percentage of marking area.

[0031] Furthermore, the classification model in step S5 employs a support vector machine or a convolutional neural network;

[0032] When using support vector machines, radial basis functions are used as kernel functions, and hyperparameters are optimized using a grid search method.

[0033] When using a convolutional neural network, the network structure includes convolutional layers, pooling layers, and fully connected layers.

[0034] Secondly, the present invention provides a computer vision-based system for sexing white-spotted beetles, comprising:

[0035] The image acquisition module is used to acquire high-resolution digital images of adult white-spotted flower beetles, including at least the dorsal, abdominal, and foretibia regions.

[0036] The image processing module is used to perform noise reduction, enhancement, and standardization processing on the acquired digital images;

[0037] The image segmentation module is used to segment the foreleg tibia region and ventral region from the preprocessed image;

[0038] The feature extraction module is used to extract morphological features from the segmented region, wherein the morphological features include at least gloss features, foreleg tibia tooth shape features, and sternite morphological features;

[0039] The classification and recognition module is used to input the extracted morphological features into a pre-trained classification model and output the sex determination result.

[0040] Furthermore, the image acquisition module includes an industrial camera, a blue ring LED light source, and an insect fixation device; the wavelength range of the blue ring LED light source is 450-470nm, and the resolution of the industrial camera is not less than 5 million pixels.

[0041] Furthermore, the system also includes a sorting execution module for sorting the insects into different collection containers based on the discrimination results.

[0042] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0043] Compared with existing technologies, this invention provides a computer vision-based method and system for sexing white-spotted flower beetles. By comprehensively utilizing multi-dimensional external morphological features of adult white-spotted flower beetles, such as stable glossiness, foreleg tibiae teeth, ventral morphology, and markings, it achieves non-destructive live identification of the insects, ensuring their subsequent utilization value in breeding and scientific research. Simultaneously, the clear process and well-defined features make operation simple and fast, significantly improving identification efficiency and reducing reliance on manual experience and specialized equipment. Furthermore, by first classifying based on glossiness and then specifically analyzing key features, the hierarchical and accurate nature of identification decisions is enhanced, providing a reliable, efficient, and low-cost solution for sex matching in large-scale artificial breeding and sex-specificity studies in wild populations. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 A flowchart illustrating the method steps provided in this embodiment of the invention;

[0046] Figure 2 This is a diagram illustrating the difference in the foretibiae of male and female white-spotted flower beetles, provided in an embodiment of the present invention.

[0047] Figure 3 This is a diagram illustrating the difference in the foretibiae of male and female *Symplocos rubrum* adults, provided in an embodiment of the present invention.

[0048] Figure 4 This is a diagram illustrating the difference in abdominal indentation between male and female adult white-spotted flower beetles provided in an embodiment of the present invention.

[0049] Figure 5This is a diagram illustrating the difference in abdominal markings between male and female adult white-spotted flower beetles provided in an embodiment of the present invention.

[0050] The image on the left shows the female insect, and the image on the right shows the male insect. Detailed Implementation

[0051] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0052] As attached Figure 1 To be continued Figure 5 As shown:

[0053] Example 1:

[0054] This invention provides a computer vision-based method for sexing white-spotted flower beetles.

[0055] The specific implementation steps are as follows:

[0056] Hardware configuration

[0057] Industrial camera: 5 megapixels, global shutter, gigabit network interface;

[0058] Lens: 8mm focal length, F1.4 aperture;

[0059] Light source: Blue ring LED light source, 150mm in diameter, 460nm wavelength, brightness continuously adjustable;

[0060] Insect fixation device: transparent stage, equipped with a miniature vacuum suction cup fixation system;

[0061] Computer: Intel i7 processor, 16GB RAM, dedicated graphics card.

[0062] Image acquisition parameter settings

[0063] Shooting distance: 12cm;

[0064] Exposure time: 5ms;

[0065] Gain: 1.5x;

[0066] Aperture: F8;

[0067] Image format: BMP, resolution 2592×1944;

[0068] Three images were collected for each insect: dorsal view, ventral view, and lateral view (showing the foretibia).

[0069] Image preprocessing

[0070] First, the acquired image undergoes non-local mean filtering for noise reduction, which removes noise while preserving edge details. Then, contrast-limited adaptive histogram equalization (CLAHE) is used for contrast enhancement. This algorithm divides the image into multiple small regions, performs histogram equalization within each region, and avoids noise amplification through contrast limiting. Finally, the image size is normalized to 1200×900 pixels for easier subsequent processing.

[0071] Region of Interest Segmentation

[0072] The specific process for foreleg tibia region segmentation is as follows: First, the RGB image is converted to the HSV color space, and the saturation channel image is extracted; then, the Otsu adaptive thresholding algorithm is used to perform binarization segmentation on the saturation channel image. The Otsu algorithm can automatically determine the optimal threshold; morphological closing operation is performed on the binarized image to fill small holes; finally, the largest connected region is extracted as the foreleg tibia region.

[0073] The web region segmentation adopts an edge detection and region growing method: first, the Canny edge detection algorithm is used to extract the web contour, and then the central region of the web is used as the seed point for region growing, finally obtaining the complete web region.

[0074] Feature extraction

[0075] (1) Gloss feature extraction:

[0076] Calculate the mean and standard deviation of each channel in the RGB color space for the dorsal region of the insect body, and calculate the gloss index G using the following formula: σ is the channel standard deviation and μ is the channel mean. The smaller the value, the smoother and shinier the insect surface.

[0077] (2) Extraction of tooth shape features of the foreleg tibia:

[0078] Canny edge detection was performed on the forefoot tibia region image to obtain the edge image. Contours were extracted from the edge image, and the three longest contours were selected as the contours of the three teeth. The centroid coordinates of each tooth contour were calculated, and the distance between the centroids of the first and second teeth was calculated using the Euclidean distance formula. The distance between the centroids of the second and third teeth .

[0079] Calculate the ratio R of the profile area of ​​the third tooth to that of the first tooth. Calculate the roundness C of the third tooth profile; the formula for roundness is... , where A is the area of ​​the outline and P is the perimeter of the outline.

[0080] Observations show that, for shiny individuals, the female insects... Usually significantly greater than Furthermore, the roundness value of the third tooth profile is small (tooth tip); the male insect's and The difference is small, and the roundness value of the third tooth profile is large (blunt tooth). For dull individuals, the tooth shape is shorter and blunter overall, but the relative difference trend between males and females still exists.

[0081] (3) Extraction of morphological features of the web:

[0082] The image of the web region is converted to grayscale. The grayscale projection curve of the web region in the vertical direction is calculated, i.e., the grayscale values ​​of each row of pixels are summed. The grayscale projection curve is smoothed using a Savitzky-Golay filter, and its first derivative is calculated. The boundary of the web concave region is determined by the zero-crossing point of the first derivative.

[0083] The indentation depth index D is calculated using the following formula: ,in The maximum gray value outside the concave area. This represents the minimum gray value at the center of the depression.

[0084] The formula for calculating the area ratio S of the depression is as follows: ,in The area of ​​the concave region. The total area of ​​the abdominal plate region is represented by D and S values, which are usually significantly larger in females than in males.

[0085] (4) Extraction of surface markings:

[0086] White spots were detected in the back region image, using pixels with a V value > 200 and an S value < 50 in the HSV color space as candidate spots. Density-based clustering analysis was performed on these candidate spots to identify the spotted regions. The number of spotted regions, N, was counted, and the ratio of the total spotted area to the area of ​​the back region was calculated. Males typically have more prominent and numerous markings (N and (The value is larger).

[0087] Based on prior classification of gloss, a gloss threshold is set (e.g., determined through statistical training set samples). If the extracted gloss feature is less than or equal to the threshold, the current insect body is classified into the "glossy" category; if the extracted gloss feature is greater than the threshold, it is classified into the "dull" category.

[0088] SVM Model Training and Classification

[0089] A training dataset was constructed containing 600 pairs of sex-labeled images of white-spotted scarab beetles (300 males and 300 females), manually labeled by entomologists. The dataset was divided into training, validation, and test sets in a 7:2:1 ratio.

[0090] Perform the above feature extraction steps on each image in the training dataset to obtain a 15-dimensional feature vector. , R, C, D, S, N (The mean and standard deviation of the three color channels, G, and the three color channels). The features were standardized using StandardScaler so that the mean of each feature was 0 and the variance was 1.

[0091] The key point is that during the training phase, this SVM model has already learned the mapping relationships between various features and gender labels in two subsets: "glossy" and "dull". During actual prediction, the model combines prior classification information on glossiness, automatically calling upon or emphasizing the corresponding discriminative knowledge, thereby outputting the final sex determination result and confidence level.

[0092] Support Vector Machine (SVM) was used as the classification model, with Radial Basis Function (RBF) chosen as the kernel function. The hyperparameters C and γ of the SVM model were optimized using a grid search method, with a search range of C = [0.1, 1, 10, 100] and γ = [0.001, 0.01, 0.1, 1]. Five-fold cross-validation was used to train the SVM model to ensure its generalization ability.

[0093] The model performance was evaluated on an independent test set (60 samples), achieving an accuracy of 96.7%, a precision (female) of 95.8%, a recall (female) of 97.1%, an F1 score of 96.4%, and a mean processing time of 3.2 seconds per animal.

[0094] The trained model is integrated into a visualization software developed using QT. The software interface includes an image display area, a parameter setting panel, a results display area, and a data management module. The system supports integrated operation of image acquisition, processing, feature extraction, classification, and result display.

[0095] Example 2:

[0096] This embodiment provides a computer vision-based method for sexing white-spotted flower beetles.

[0097] Network architecture design

[0098] The network architecture uses a lightweight CNN design, specifically including:

[0099] Input layer: Receives 224×224×3 RGB images;

[0100] The first convolutional block contains two convolutional layers, each using 64 3×3 convolutional kernels, batch normalization, ReLU activation function, and 2×2 max pooling;

[0101] The second convolutional block contains two convolutional layers, each using 128 3×3 convolutional kernels, batch normalization, ReLU activation function, and 2×2 max pooling.

[0102] The third convolutional block contains 3 convolutional layers, each using 256 3×3 convolutional kernels, batch normalization, ReLU activation function, and global average pooling;

[0103] Fully connected layer: 512 neurons, Dropout rate 0.5;

[0104] Output layer: 2 neurons, with Softmax activation function, corresponding to the probabilities of female and male respectively.

[0105] Data augmentation strategy

[0106] To increase the diversity of training data and prevent overfitting, the following data augmentation methods are used:

[0107] Random rotation: Rotate the image randomly within a range of ±15 degrees;

[0108] Horizontal flip: Perform a horizontal flip with a probability of 0.5;

[0109] Brightness adjustment: Adjust image brightness within a range of ±20%;

[0110] Contrast Adjustment: Adjust image contrast within a range of ±15%;

[0111] Add Gaussian noise: Add Gaussian noise with a standard deviation of 0.01.

[0112] Training process

[0113] The Adam optimizer was used with an initial learning rate of 0.001. Early stopping was employed to prevent overfitting; training was stopped when the validation set accuracy did not improve for 15 consecutive epochs. A learning rate decay strategy was used, halving the learning rate when the validation loss did not decrease for 5 consecutive epochs. Training was conducted for 100 epochs with a batch size of 32.

[0114] Performance evaluation

[0115] Performance of the CNN model on the same test set:

[0116] Accuracy: 97.8% (better than SVM's 96.7%).

[0117] Accuracy (female): 97.2%;

[0118] Recall rate (female): 98.3%;

[0119] F1 score: 97.7%;

[0120] Average processing time: 1.8 seconds per device (GPU accelerated).

[0121] Model Deployment

[0122] The trained CNN model is converted to TensorFlow Lite format and deployed to embedded devices (such as Jetson Nano) to achieve edge computing. The system can run in an environment without a network connection, protecting data privacy and reducing latency.

[0123] Example 3:

[0124] This embodiment provides a computer vision-based system for sexing white-spotted flower beetles.

[0125] The system hardware includes:

[0126] Feeding mechanism: Vibratory feeder system, which transports the insects to the shooting station in an orderly manner;

[0127] Image acquisition station: Three-camera system (simultaneous shooting from top, bottom, and side), with a camera resolution of 5 megapixels;

[0128] Sorting mechanism: a three-axis robotic arm with a negative pressure suction nozzle, with a positioning accuracy of ±0.5mm;

[0129] Collection containers: Male and female insects are collected separately, and each container is equipped with an infrared counting sensor;

[0130] Control cabinet: industrial PC, PLC controller, power supply module.

[0131] The software system adopts a layered architecture design:

[0132] Equipment control layer: Controls hardware devices such as cameras, light sources, and robotic arms via the Modbus TCP protocol;

[0133] Image processing layer: Implements image preprocessing, segmentation, and feature extraction algorithms based on OpenCV;

[0134] Intelligent decision-making layer: Runs the trained classification model to distinguish between male and female;

[0135] Human-computer interaction layer: A graphical interface developed based on the Qt framework, displaying real-time status, historical data, and statistical reports.

[0136] The system workflow is as follows:

[0137] The insect body is transported to the shooting station by the feeding mechanism and positioned by vibration and airflow.

[0138] The insect was fixed in place by a vacuum suction cup, and three cameras simultaneously captured images of its back, abdomen, and side views.

[0139] The image processing module performs fusion analysis on the three views and extracts multi-dimensional features;

[0140] The classification and recognition module determines male or female based on the extracted features and outputs the results and confidence scores.

[0141] The robotic arm sorts the insects into the corresponding collection containers based on the identification results;

[0142] The system automatically records the image, feature data, identification results, and timestamp of each insect.

[0143] System performance testing

[0144] The system ran continuously for 8 hours, processing 5760 insects.

[0145] Average accuracy: 96.3%;

[0146] Highest instantaneous error rate: 4.2%;

[0147] Number of system failures: 0;

[0148] Peak memory usage: 2.3GB;

[0149] Processing capacity: 720 pieces / hour;

[0150] Insect damage rate: <0.1%;

[0151] Average energy consumption: 350W.

[0152] Adaptability test

[0153] Tests were conducted on white-spotted flower beetles of different instars:

[0154] Insect age 1-7 days: accuracy rate 94.0% (sample size 50);

[0155] Insect age 8-14 days: accuracy rate 97.2% (sample size 50);

[0156] Insect age 15-21 days: accuracy rate 96.8% (sample size 50);

[0157] Insect age 22-30 days: accuracy rate 95.6% (sample size 50).

[0158] Experimental results show that the method of the present invention maintains a high identification accuracy rate in different insect age groups and has strong adaptability.

[0159] Technical effect verification

[0160] To verify the technical effects of the present invention, the following comparative experiments were conducted:

[0161] Experiment 1: Accuracy Comparison

[0162] Identification methods Sample size accuracy Processing time (seconds / piece) manual observation 100 84.0% 38.5 Anatomical identification 100 100% 65.2 This invention (SVM) 100 96.7% 3.2 This invention (CNN) 100 97.8% 1.8

[0163] Experiment 2: Comparison of Effects from Different Light Sources

[0164] Light source type Wavelength (nm) accuracy Feature contrast White LED 400-700 92.3% medium Blue LED 450-470 96.7% high Red LED 620-660 88.5% Low

[0165] Experiment 3: The Effect of Different Resolutions

[0166] Camera resolution accuracy Processing time Storage space (single image) 2 megapixels 93.2% 2.1s 1.2MB 5 megapixels 96.7% 3.2s 3.0MB 12 megapixels 97.1% 5.8s 7.2MB

[0167] This invention, through the deep integration of computer vision technology and insect morphology, achieves automated, intelligent, and standardized sex identification of the white-spotted flower beetle. It has promising application prospects and promotional value, and can be widely applied in: large-scale white-spotted flower beetle farms for seed selection and sex ratio control to improve reproductive efficiency; biological control engineering for pest monitoring and control strategy formulation; research institutions for insect behavioral and ecological research; quarantine and inspection departments for monitoring and controlling invasive species; and insect resource development and utilization for the breeding and management of medicinal insects and protein feed insects.

[0168] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for sexing white-spotted flower beetles based on computer vision, characterized in that, include: S1. Obtain high-resolution digital images of adult white-spotted flower beetles, including at least the dorsal, abdominal, and fore tibia regions; S2. Denoise, enhance, and standardize the acquired digital images; S3. Segment the foreleg tibia region and ventral region from the preprocessed image; S4. Extract morphological features from the segmented region, wherein the morphological features include at least gloss features, foreleg tibia tooth shape features, and sternite morphological features; S5. Input the extracted morphological features into the pre-trained classification model and output the sex determination result.

2. The method for sexing white-spotted flower beetles based on computer vision according to claim 1, characterized in that, In step S4, the gloss feature is obtained by calculating the statistical features of the color channel in the dorsal region of the insect body, and is used to classify the insect body into a glossy category or a matte category.

3. The method for sexing white-spotted flower beetles based on computer vision according to claim 1, characterized in that, In step S4, the extraction of the tooth shape features of the forefoot tibia includes the following sub-steps: Edge detection was performed on the forefoot tibia region image to extract contour features; The outlines of the three teeth are selected, and the centroid coordinates of the outline of each tooth are calculated. Calculate the distance between the centroids of the first and second teeth, and the distance between the centroids of the second and third teeth; Calculate the ratio of the contour area of ​​the third tooth to that of the first tooth; Calculate the roundness of the third tooth profile; Distance, area ratio, and roundness were used as characteristics of the forefoot tibia teeth.

4. The method for sexing white-spotted flower beetles based on computer vision according to claim 1, characterized in that, In step S4, the extraction of the web morphological features includes the following sub-steps: The image of the abdominal region is converted to grayscale. Calculate the grayscale projection curve of the web region in the vertical direction; The grayscale projection curve is smoothed and its first derivative is calculated to determine the web concave region. The depth of the depression and the percentage of the depression area are calculated as morphological characteristics of the web.

5. The method for sexing white-spotted flower beetles based on computer vision according to claim 1, characterized in that, In step S4, the extracted morphological features also include surface markings. The surface markings were obtained through white marking detection and cluster analysis, including the number of markings and the percentage of marking area.

6. The method for sexing white-spotted flower beetles based on computer vision according to claim 1, characterized in that, The classification model in step S5 uses a support vector machine or a convolutional neural network. When using support vector machines, radial basis functions are used as kernel functions, and hyperparameters are optimized using a grid search method. When using a convolutional neural network, the network structure includes convolutional layers, pooling layers, and fully connected layers.

7. A computer vision-based system for sexing white-spotted beetles, characterized in that, include: The image acquisition module is used to acquire high-resolution digital images of adult white-spotted flower beetles, including at least the dorsal, abdominal, and foretibia regions. The image processing module is used to perform noise reduction, enhancement, and standardization processing on the acquired digital images; The image segmentation module is used to segment the foreleg tibia region and ventral region from the preprocessed image; The feature extraction module is used to extract morphological features from the segmented region, wherein the morphological features include at least gloss features, foreleg tibia tooth shape features, and sternite morphological features; The classification and recognition module is used to input the extracted morphological features into a pre-trained classification model and output the sex determination result.

8. A computer vision-based sexing system for white-spotted flower beetles according to claim 7, characterized in that, The image acquisition module includes an industrial camera, a blue ring LED light source, and an insect fixation device; the wavelength range of the blue ring LED light source is 450-470nm, and the resolution of the industrial camera is no less than 5 million pixels.

9. A computer vision-based sexing system for white-spotted flower beetles according to claim 7, characterized in that, The system also includes a sorting execution module, which sorts the insects into different collection containers based on the discrimination results.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method steps as described in any one of claims 1-6.

Images