Ginkgo male and female identification method and system based on RGB image background expansion
By combining RGB image background augmentation and deep learning technology with multiple network models to identify the sex of ginkgo trees, the problem of high false positive rate and high cost in existing methods is solved, and a fast, accurate and low-cost method for identifying the sex of ginkgo trees is achieved.
Patent Information
- Application Number
- CN202311154970.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-07
- Publication Date
- 2025-12-26
- Estimated Expiration
- 2043-09-07
AI Technical Summary
Existing methods for identifying the sex of ginkgo trees rely heavily on experience, lack clear digital and graphical standards, resulting in a high misjudgment rate. Furthermore, these methods are complex, costly, and difficult to promote on a large scale.
Using RGB image background augmentation technology, images of ginkgo leaves are captured by an RGB camera, and then binarized and augmented. Combined with Resnet101, Inception-v3, and Inception-Resnet-v2 network models, a majority voting decision is used to form a ginkgo sex classification model, achieving fast and accurate sex identification.
It enables rapid, accurate, and low-cost detection for gender identification in ginkgo, lowers the technical threshold, has a wide range of applications, and achieves a detection accuracy of over 91%, solving the problems of complexity and high cost of existing methods.
Smart Images
Figure CN117237710B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of forestry production, in particular to a ginkgo male and female identification method and system based on RGB image background expansion. BACKGROUND
[0002] Ginkgo is a plant of Ginkgoaceae, which first appeared in the Carboniferous period 345 million years ago, and has undergone great changes for more than 300 million years. It is the oldest surviving plant in the world and is recognized as a "living fossil". In addition to historical and cultural value, ginkgo has multiple functions such as timber, fruit, medicine, greening, and beautification in actual production and life. However, ginkgo has the characteristics of dioecy, and male and female plants have different values in practical application. Male plants have higher ornamental value due to their height, dense crown, and long leaf period, while female plants produce seeds that naturally fall off after maturation, increasing sanitation costs, and the outer seed coat emits a foul odor after being crushed. Therefore, male plants have more advantages in landscape greening. On the other hand, ginkgo seeds have physiological and pharmacological activities such as ginkgolic acid, making them a product with high economic value. In addition, female plants have higher value than male plants in terms of leaf value. Therefore, ginkgo economic forests are mainly female plants. Different planting needs require ginkgo male and female plants to be planted separately, so it is of great significance to establish a fast, accurate, and low-cost ginkgo male and female identification method.
[0003] The simplest and most intuitive method to distinguish gender is through the morphological characteristics of ginkgo male and female plants, but this method has high dependence on experience, insufficient reliability, lack of digitalization and clear atlas standards, large differences in shape and quality indicators, and is prone to misjudgment. SUMMARY
[0004] The purpose of the present application is to provide a ginkgo male and female identification method and system based on RGB image background expansion to solve the above problems.
[0005] To achieve the above purpose, the present application provides the following solutions:
[0006] In a first aspect, the present application provides a ginkgo male and female identification method based on RGB image background expansion, comprising:
[0007] Obtain ginkgo leaves of different genders at different stages as sample ginkgo leaves, and use an RGB camera to take pictures of the sample ginkgo leaves to obtain sample ginkgo leaf RGB images;
[0008] Perform a binary segmentation operation on all sample ginkgo leaf RGB images to obtain sample ginkgo leaf images without background, and perform background expansion on the sample ginkgo leaf images without background to obtain background-expanded sample ginkgo leaf images;
[0009] A true label is marked for each background-extended sample ginkgo leaf image to obtain sample data; wherein the true label is male or female;
[0010] The sample data is divided into a modeling set, a verification set and a prediction set according to a certain proportion by using random sampling, and the data in the modeling set and the verification set are used to train a Resnet101 network model, an Inception-v3 network model and an Inception-Resnet-v2 network model respectively to obtain three ginkgo male and female classification models;
[0011] The data in the prediction set are input into the three determined male and female classification models respectively to determine the male and female classification prediction results of each male and female classification model, and the three male and female classification models are packaged to form a majority decision ginkgo male and female prediction model according to the male and female classification prediction results of each male and female classification model by using majority voting decision.
[0012] The RGB image of the ginkgo leaf to be detected is input into the majority decision ginkgo male and female prediction model to obtain the male and female classification identification result of the ginkgo leaf.
[0013] Optionally, ginkgo leaves of different genders at different periods are obtained as sample ginkgo leaves, and the sample ginkgo leaves are photographed by using an RGB camera to obtain sample ginkgo leaf RGB images, which specifically include:
[0014] No less than 1000 ginkgo leaves of different genders at different periods are collected as sample ginkgo leaves, and then the sample ginkgo leaves are wiped to remove surface stains and laid flat on a black light-absorbing cloth, and the sample ginkgo leaves are photographed by using an RGB camera in a manner perpendicular to the sample ginkgo leaves to obtain sample ginkgo leaf RGB images.
[0015] Optionally, a binaryzation segmentation operation is performed on all sample ginkgo leaf RGB images, which specifically includes:
[0016] An Otsu algorithm is used to perform a binaryzation segmentation operation on all sample ginkgo leaf RGB images.
[0017] Optionally, the sample ginkgo leaf images without background are background-extended to obtain background-extended sample ginkgo leaf images, which specifically include:
[0018] An equal proportion image size adjustment method is used to adjust the long side of the sample ginkgo leaf image without background to a set value, and black is filled around the image, and then the sample ginkgo leaf image without background is adjusted to a square image with the adjusted long side as the side length of the square; the square image is the background-extended sample ginkgo leaf image.
[0019] Optionally, the Resnet101 network model, the Inception-v3 network model and the Inception-Resnet-v2 network model are trained respectively by using the data in the modeling set and the verification set, and three ginkgo male and female classification models are obtained, specifically including:
[0020] The modeling set and the verification set are subjected to data enhancement processing.
[0021] The Resnet101 network model is trained by using the modeling set and the verification set subjected to data enhancement processing, and a first ginkgo male and female classification model is obtained.
[0022] The Inception-v3 network model is trained by using the modeling set and the verification set subjected to data enhancement processing, and a second ginkgo male and female classification model is obtained.
[0023] The Inception-Resnet-v2 network model is trained by using the modeling set and the verification set subjected to data enhancement processing, and a third ginkgo male and female classification model is obtained.
[0024] Optionally, the data enhancement processing includes image rotation and image scaling; wherein the data enhancement processing mode of the modeling set is the same as that of the verification set.
[0025] In the second aspect, the application provides a ginkgo male and female identification system based on RGB image background expansion, comprising:
[0026] A sample ginkgo leaf RGB image acquisition module is configured to acquire ginkgo leaves of different genders at different periods as sample ginkgo leaves, and to capture the sample ginkgo leaves by using an RGB camera to obtain sample ginkgo leaf RGB images.
[0027] A background expansion module is configured to perform a binaryzation segmentation operation on all sample ginkgo leaf RGB images to obtain sample ginkgo leaf images without background, and to perform background expansion on the sample ginkgo leaf images without background to obtain sample ginkgo leaf images subjected to background expansion.
[0028] A sample data determination module is configured to mark a real label for each sample ginkgo leaf image subjected to background expansion to obtain sample data; wherein the real label is male or female.
[0029] A training module is configured to divide the sample data into a modeling set, a verification set and a prediction set according to a certain proportion by using a random sampling method, and to train a Resnet101 network model, an Inception-v3 network model and an Inception-Resnet-v2 network model respectively by using the data in the modeling set and the verification set to obtain three ginkgo male and female classification models.
[0030] The majority decision ginkgo male and female prediction model determination module is configured to input data in the prediction set into the three determined male and female classification models respectively, determine male and female classification prediction results of each male and female classification model, and encapsulate the three determined male and female classification models by majority voting decision to form a majority decision ginkgo male and female prediction model.
[0031] The ginkgo leaf male and female identification module is configured to input an RGB image of a ginkgo leaf to be detected into the majority decision ginkgo male and female prediction model to obtain a male and female classification identification result of the ginkgo leaf.
[0032] According to the specific embodiments provided by the present application, the following technical effects are disclosed.
[0033] The present application takes the RGB image of the ginkgo leaf as the detection target, and realizes rapid, accurate and low-cost detection of the ginkgo gender by combining the deep learning technology, solves the problems of complex operation, long time consumption, high cost and high technical threshold of other ginkgo gender identification methods, and provides technical support for actual ginkgo planting and production. BRIEF DESCRIPTION OF DRAWINGS
[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0035] Figure 1 The flowchart of the ginkgo male and female identification method based on RGB image background expansion provided by the embodiment of the present application is shown.
[0036] Figure 2 The image background expansion schematic diagram provided by the embodiment of the present application is shown.
[0037] Figure 3 The overall flowchart of the ginkgo male and female identification method based on RGB image background expansion provided by the embodiment of the present application is shown.
[0038] Figure 4 The comparison result diagram of the image size adjustment mode provided by the embodiment of the present application is shown.
[0039] Figure 5 The influence schematic diagram of the image size adjustment mode provided by the embodiment of the present application is shown.
[0040] Figure 6 The ginkgo leaf identification result interface diagram provided by the embodiment of the present application is shown. DETAILED DESCRIPTION
[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0043] Example 1
[0044] This embodiment provides a method for identifying the sex of ginkgo trees based on RGB image background augmentation, such as... Figure 1 As shown, it includes:
[0045] Step 100: Obtain ginkgo leaves of different ages and genders as sample ginkgo leaves, and take pictures of the sample ginkgo leaves with an RGB camera to obtain RGB images of the sample ginkgo leaves.
[0046] In this embodiment, no less than 1,000 ginkgo leaves of different ages and sexes are collected as sample ginkgo leaves. After wiping off surface dirt, the sample ginkgo leaves are laid flat on a black light-absorbing cloth. An RGB camera is used to take pictures of the sample ginkgo leaves in a perpendicular manner to the sample ginkgo leaves to obtain RGB images of the sample ginkgo leaves. The distance between the RGB camera and the sample ginkgo leaves is 10-30cm.
[0047] Step 200: Perform binarization segmentation on all sample ginkgo leaf RGB images to obtain sample ginkgo leaf images without background, and perform background augmentation on the sample ginkgo leaf images without background to obtain sample ginkgo leaf images with augmented background.
[0048] In this embodiment, the Otsu algorithm is used to perform binarization segmentation on all sample ginkgo leaf RGB images.
[0049] In this embodiment, as Figure 2 As shown, the background expansion process is as follows:
[0050] By proportionally adjusting the image size, the long side of the sample ginkgo leaf image without background is adjusted to the set value, and black is filled around the image. Then, the sample ginkgo leaf image without background is adjusted to a square image with the adjusted long side as the side length of the square. At this time, the square image is the sample ginkgo leaf image with background expansion.
[0051] Step 300: labeling a real label for each background-expanded sample ginkgo leaf image to obtain sample data; wherein the real label is male or female.
[0052] Step 400: using random sampling method, dividing the sample data into modeling set, verification set and prediction set according to a certain proportion, and using the data in the modeling set and the verification set to train Resnet101 network model, Inception-v3 network model and Inception-Resnet-v2 network model respectively to obtain three ginkgo male and female classification models.
[0053] In this embodiment, step 400 specifically comprises:
[0054] (1) data augmentation processing is performed on the modeling set and the verification set; the data augmentation processing includes image rotation and image scaling; wherein the data augmentation processing mode of the modeling set is the same as that of the verification set.
[0055] (2) using the modeling set and the verification set after data augmentation processing, the Resnet101 network model is trained to obtain the first ginkgo male and female classification model.
[0056] (3) using the modeling set and the verification set after data augmentation processing, the Inception-v3 network model is trained to obtain the second ginkgo male and female classification model.
[0057] (4) using the modeling set and the verification set after data augmentation processing, the Inception-Resnet-v2 network model is trained to obtain the third ginkgo male and female classification model.
[0058] Further, in the model training process, the loss value of the classification result belonging to the classification result and the real label of the sample is calculated by the loss function, which is fed back to the deep convolutional neural network to adjust the model parameters, and one training process of the sample data is completed. The training set sample data is divided into not less than 16 batches, the training times reach 150 times or the loss value does not decrease for 25 consecutive times, and the three ginkgo male and female classification models are obtained.
[0059] Step 500: inputting the data in the prediction set into the three determined male and female classification models respectively, determining the male and female classification prediction results of each male and female classification model, and according to the male and female classification prediction results of each male and female classification model, using majority voting decision to encapsulate the three determined male and female classification models to form a majority decision ginkgo male and female prediction model.
[0060] Step 600: inputting the RGB image of the ginkgo leaf to be detected into the majority decision ginkgo male and female prediction model to obtain the male and female classification identification result of the ginkgo leaf.
[0061] The overall flowchart of the embodiment can be seen from Figure 3 .
[0062] The device involved in the embodiment only includes an RGB camera, without other expensive instruments, and has the advantages of simple and easy-to-obtain device and low cost. Meanwhile, the identification process is simple to operate and has low technical threshold, which can solve the problem that the current common ginkgo male and female identification methods based on molecular markers and physiological and biochemical characteristics have high technical barriers and cannot be widely promoted.
[0063] Embodiment Two
[0064] The ginkgo male and female identification method based on RGB image background expansion proposed in the embodiment will be described in further detail below with reference to the accompanying drawings, and the specific steps include:
[0065] Step 1: On May 11 and July 2, green leaf period ginkgo leaves were collected, a total of 1271 pieces, of which 635 pieces were male tree leaves. On November 2, yellow leaf period ginkgo leaves were collected, a total of 1306 pieces, of which 652 pieces were male tree leaves.
[0066] Step 2: After wiping off the surface stains of the ginkgo leaves, the ginkgo leaves were naturally laid on a black light-absorbing cloth, the RGB camera was perpendicular to the ginkgo leaves, the distance between the RGB camera and the ginkgo leaves was 15 cm, and the photos were taken.
[0067] Step 3: Otsu algorithm was used to perform binary segmentation on the RGB image of the ginkgo leaves, and the ginkgo leaf image without background was obtained. The image size was adjusted in proportion, the long side of the image was 560, the peripheral area of the image was filled with black, a 560*560 size image was formed, and the authenticity gender label was marked. The image size adjustment method had an effect on the results of the Resnet101 model, as shown in Figure 4 . Figure 5 .
[0068] Step 4: The random sampling method was used to divide the obtained images into modeling set, verification set and prediction set according to the ratio of 3:1:1.
[0069] Step 5: The modeling set and the verification set were subjected to data enhancement processing, and the enhancement processing method was rotation and scaling. The rotation range was -90°-90°, and an enhanced image was obtained every 10°. The scaling range was 0.5-2 times, and an enhanced image was obtained every 0.1 times, i.e. the data set was expanded by 33 times.
[0070] Step 6: The data set after data augmentation is input into the Resnet101, Inception-v3, and Inception-Resnet-v2 network models. The image size of the input Resnet101 network model is uniformly converted to 224*224, and the image size of the Inception-v3 and Inception-Resnet-v2 network models is uniformly converted to 299*299.
[0071] Step 7: The initial parameters of the above three deep learning models are the results obtained by pre-training on the ImageNet data set. Cross-entropy is selected as the loss function, SGDM is selected as the optimizer, the iteration number is set to 150, the batch processing parameter is set to 64, the learning rate annealing strategy is adopted, the initial learning rate is 0.01, the learning rate is reduced by 10% after every 30 iterations, and the parameters with the highest accuracy on the validation set are saved as the trained model. The accuracy of the trained model on the validation set is 91.84% for the Resnet101 network model, 89.13% for the Inception-v3 network model, and 89.13% for the Inception-Resnet-v2 network model.
[0072] Step 8: The prediction set image is input into the three established male and female classification models, and the accuracy is 91.44% for the Resnet101 network model, 89.69% for the Inception-v3 network model, and 88.33% for the Inception-Resnet-v2 network model. The resulting models are packaged to form a majority decision ginkgo male and female prediction model, and the accuracy of this model on the prediction set is 92.02%.
[0073] Step 9: A female ginkgo plant leaf is input into the majority decision ginkgo male and female prediction model, and the gender identification result is female, as shown in Figure 6
[0074] Example Three
[0075] In order to perform the method corresponding to the above-mentioned example one, to realize the corresponding functions and technical effects, a ginkgo male and female identification system based on RGB image background expansion is provided.
[0076] The ginkgo male and female identification system based on RGB image background expansion provided in this embodiment comprises:
[0077] A sample ginkgo leaf RGB image acquisition module is configured to acquire ginkgo leaves of different genders at different times as sample ginkgo leaves, and to capture the sample ginkgo leaves using an RGB camera to obtain sample ginkgo leaf RGB images.
[0078] The background expansion module is configured to perform a binaryzation segmentation operation on all sample ginkgo leaf RGB images to obtain sample ginkgo leaf images without background, and to perform background expansion on the sample ginkgo leaf images without background to obtain sample ginkgo leaf images after background expansion.
[0079] The sample data determination module is configured to mark a true label for each sample ginkgo leaf image after background expansion to obtain sample data, wherein the true label is male or female.
[0080] The training module is configured to divide the sample data into a modeling set, a verification set and a prediction set according to a certain proportion by using a random sampling method, and to train Resnet101 network model, Inception-v3 network model and Inception-Resnet-v2 network model respectively by using data in the modeling set and the verification set to obtain three ginkgo male and female classification models.
[0081] The majority decision ginkgo male and female prediction model determination module is configured to input data in the prediction set into the three determined male and female classification models respectively, to determine male and female classification prediction results of each male and female classification model, and to encapsulate the three determined male and female classification models to form a majority decision ginkgo male and female prediction model according to the male and female classification prediction results of each male and female classification model by using majority voting decision.
[0082] The ginkgo leaf male and female identification module is configured to input an RGB image of a ginkgo leaf to be detected into the majority decision ginkgo male and female prediction model to obtain a male and female classification identification result of the ginkgo leaf.
[0083] Embodiment four
[0084] The embodiment of the present application provides an electronic device including a memory for storing a computer program and a processor for running the computer program to enable the electronic device to perform the ginkgo male and female identification method based on RGB image background expansion of the embodiment one.
[0085] Optionally, the electronic device can be a server.
[0086] In addition, the embodiment of the present application further provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the ginkgo male and female identification method based on RGB image background expansion of the embodiment one.
[0087] Compared with the prior art, the present application has the following advantages:
[0088] (1) The device is simple and low in cost.
[0089] The device related to the application only comprises an RGB camera, without other expensive instruments, so that the device is simple and easy to obtain, and has the advantages of low cost.
[0090] (2) Simple operation, low technical threshold
[0091] The application only needs to take a picture of ginkgo leaves and put the picture into an established model to obtain the gender classification result, and has the advantages of simple operation and low technical threshold. Compared with the commonly used ginkgo sex identification method based on molecular markers and physiological and biochemical characteristics, the method involves complex operations such as gene amplification and electrochemical determination, and has high technical threshold, so it cannot be popularized on a large scale.
[0092] (3) Wide application range and high accuracy.
[0093] The model contained in the detection method is trained based on the whole year growth cycle data samples of ginkgo leaves, so that the sex identification of ginkgo at any time can be realized. Meanwhile, the detection accuracy of the application is high, and after multiple verifications, the overall accuracy can reach more than 91%.
[0094] In the specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts of each embodiment can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the related parts can be referred to the method part.
[0095] In the present application, specific examples are applied to the principle and implementation mode of the application, and the above embodiment is only used to help understand the method and core idea of the application; at the same time, for the general technical personnel in the art, according to the idea of the application, the specific implementation mode and application range will be changed. In view of the above, the content of the specification should not be understood as the limitation of the application.
Claims
1. A method for identifying the sex of ginkgo trees based on RGB image background augmentation, characterized in that, include: Ginkgo leaves from different periods and of different sexes were collected as sample ginkgo leaves, and RGB images of the sample ginkgo leaves were obtained by taking pictures of the sample ginkgo leaves with an RGB camera. Binarization segmentation was performed on all sample ginkgo leaf RGB images to obtain sample ginkgo leaf images without background. Background augmentation was then performed on the sample ginkgo leaf images without background to obtain sample ginkgo leaf images with background augmentation. Each ginkgo leaf image after background augmentation was labeled with a real label to obtain sample data; where the real label is male or female. Using a random sampling method, the sample data was divided into a modeling set, a validation set, and a prediction set according to a certain ratio. The Resnet101 network model, the Inception-v3 network model, and the Inception-Resnet-v2 network model were trained using the data from the modeling set and the validation set, respectively, to obtain three Ginkgo sex classification models. The data in the prediction set are input into the three determined male and female classification models respectively, the male and female classification prediction results of each male and female classification model are determined, and the three determined male and female classification models are encapsulated by majority voting based on the male and female classification prediction results of each male and female classification model to form the majority decision ginkgo male and female prediction model. The RGB images of the ginkgo leaves to be tested are input into the majority decision ginkgo sex prediction model to obtain the sex classification and identification results of the ginkgo leaves.
2. The method for identifying the sex of ginkgo trees based on RGB image background augmentation according to claim 1, characterized in that, Ginkgo leaves from different periods and of different sexes were collected as sample ginkgo leaves, and RGB images of the sample ginkgo leaves were obtained using an RGB camera. Specifically, the images include: Collect no fewer than 1,000 ginkgo leaves of different ages and sexes as sample ginkgo leaves. After wiping off surface dirt, lay the sample ginkgo leaves flat on a black light-absorbing cloth. Use an RGB camera to photograph the sample ginkgo leaves perpendicular to them to obtain RGB images of the sample ginkgo leaves.
3. The method for identifying the sex of ginkgo trees based on RGB image background augmentation according to claim 1, characterized in that, Binarization segmentation was performed on all sample Ginkgo leaf RGB images, specifically including: The Otsu algorithm was used to perform binarization segmentation on all sample Ginkgo leaf RGB images.
4. The method for identifying the sex of ginkgo trees based on RGB image background augmentation according to claim 1, characterized in that, Background augmentation is performed on sample ginkgo leaf images without background to obtain background-augmented sample ginkgo leaf images, specifically including: By proportionally adjusting the image size, the long side of the backgroundless sample ginkgo leaf image is adjusted to a set value, and black is filled around the image. Then, the backgroundless sample ginkgo leaf image is adjusted into a square image with the adjusted long side as the side length of the square; the square image is the sample ginkgo leaf image with the background expanded.
5. The method for identifying the sex of ginkgo trees based on RGB image background augmentation according to claim 1, characterized in that, Using data from the modeling and validation sets, ResNet101, Inception-v3, and Inception-ResNet-v2 network models were trained respectively, resulting in three Ginkgo sex classification models, specifically including: Perform data augmentation on the modeling and validation sets; Using the data augmented modeling and validation sets, a ResNet101 network model was trained to obtain the first Ginkgo sex classification model. Using the data-augmented modeling and validation sets, the Inception-v3 network model was trained to obtain a second ginkgo sex classification model. Using the data augmented modeling and validation sets, the Inception-Resnet-v2 network model was trained to obtain the third Ginkgo sex classification model.
6. The method for identifying the sex of ginkgo trees based on RGB image background augmentation according to claim 5, characterized in that, Data augmentation includes image rotation and image scaling; the data augmentation methods for the modeling set are the same as those for the validation set.
7. A system for identifying the sex of ginkgo trees based on RGB image background augmentation, characterized in that, include: The sample ginkgo leaf RGB image acquisition module is used to acquire ginkgo leaves of different periods and genders as sample ginkgo leaves, and to take pictures of the sample ginkgo leaves with an RGB camera to obtain RGB images of the sample ginkgo leaves. The background augmentation module is used to perform binarization segmentation on all sample ginkgo leaf RGB images to obtain sample ginkgo leaf images without background, and to augment the background of the sample ginkgo leaf images without background to obtain sample ginkgo leaf images with augmented background. The sample data determination module is used to label each background-enlarged sample Ginkgo leaf image with a real label to obtain sample data; where the real label is male or female. The training module is used to divide the sample data into a modeling set, a validation set, and a prediction set according to a certain ratio using a random sampling method. The Resnet101 network model, the Inception-v3 network model, and the Inception-Resnet-v2 network model are trained using the data in the modeling set and the validation set, respectively, to obtain three Ginkgo male and female classification models. The majority decision-making Ginkgo sex prediction model determination module is used to input the data in the prediction set into the three determined sex classification models, determine the sex classification prediction result of each sex classification model, and encapsulate the three determined sex classification models by majority voting based on the sex classification prediction results of each sex classification model to form the majority decision-making Ginkgo sex prediction model. The Ginkgo leaf sex identification module is used to input the RGB image of the Ginkgo leaf to be tested into the majority decision Ginkgo sex prediction model to obtain the sex classification and identification results of the Ginkgo leaf.