Bumblebee pollination effect detection method, device, equipment, medium and product

Abstract

The invention provides a bumblebee pollination effect detection method, device and equipment, a medium and a product. The method comprises the following steps: acquiring a to-be-detected image; the to-be-detected image comprises flowers of crops; inputting the detection image into a trained bumblebee pollination mark recognition model, so that the bumblebee pollination mark recognition model performs image recognition on each flower in the to-be-detected image to obtain a pollination mark of each flower output by the bumblebee pollination mark recognition model; and calculating the pollination success rate according to the number of the pollination marks and the number of all the flowers. The automatic detection process of the crop pollination success rate is realized by utilizing bumblebee pollination marks, the evaluation efficiency of the bumblebees on the crop pollination effect is improved, time and labor consumption and subjective deviation of manual observation are avoided, and the method is flexible, simple, convenient and efficient, saves human resources, reduces the cost and can be popularized in a large scale.

Description

Technical Field

[0001] This invention relates to the field of crop cultivation technology, and in particular to a method, apparatus, equipment, medium, and product for detecting bumblebee pollination effectiveness. Background Technology

[0002] Bumblebees are indispensable pollinating insects in the ecosystem, playing a vital role in maintaining biodiversity and promoting agricultural economic development. Their unique living habits and cold resistance make them "pollination masters" in nature, and they are key pollinators for many wild plants and crops (such as tomatoes and blueberries). They improve pollination efficiency through "vibration pollination" (releasing pollen by vibrating their wings).

[0003] Bumblebee pollination directly impacts crop fruit set, yield, and fruit quality. Therefore, accurately and quickly assessing bumblebee pollination effectiveness is crucial for optimizing pollination programs, planting management, yield prediction, and guiding agricultural production. Currently, methods for assessing bumblebee pollination intensity and effectiveness mainly include manually observing flowers for traces left by bumblebees, statistically analyzing fruit set rates, or examining pollen tube germination by dissecting flowers. However, manual observation is time-consuming, labor-intensive, and highly subjective; statistically analyzing fruit set rates is lagging and cannot promptly reflect problems in the pollination process during the current season; while pollen tube detection improves accuracy, it is destructive, complex, and unsuitable for large-scale monitoring.

[0004] It is evident that existing methods for detecting bumblebee pollination effectiveness are inefficient and difficult to promote on a large scale. Summary of the Invention

[0005] This invention provides a method, apparatus, equipment, medium, and product for detecting bumblebee pollination effectiveness, which solves the defects of existing bumblebee pollination effectiveness detection methods that are time-consuming, labor-intensive, or complex to operate, and realizes an efficient, simple bumblebee pollination effectiveness detection method that can help adjust planting plans in a timely manner.

[0006] This invention provides a method for detecting the pollination effect of bumblebees, comprising the following steps.

[0007] Acquire an image to be detected; the image to be detected includes the flowers of a crop; The detected image is input into a bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model can perform image recognition for each flower in the image to be detected, and obtain the pollination mark of each flower output by the bumblebee pollination mark recognition model. The pollination success rate is calculated based on the number of pollination markers and the total number of flowers. The bumblebee pollination marker recognition model that has been trained is based on flower sample images of the crop and pollination marker sample labels in the flower sample images of the crop.

[0008] According to a method for detecting bumblebee pollination effects provided by the present invention, the trained bumblebee pollination mark recognition model includes a trained first image segmentation model and a trained second image detection model; the detection image is input into the trained bumblebee pollination mark recognition model, so that the bumblebee pollination mark recognition model performs image recognition on each flower in the image to be detected, and obtains the pollination mark of each flower output by the bumblebee pollination mark recognition model, including: The detected image is input into the first image segmentation model that has been trained, and a flower segmentation image containing flower region markers is obtained from the output of the first image segmentation model; The flower segmentation image is input into the trained second image detection model to obtain the authorization label for the flower in the flower region label output by the second image detection model.

[0009] According to a method for detecting bumblebee pollination effect provided by the present invention, before acquiring the image to be detected, the method further includes: Obtain a first set of flower sample images; each flower sample image in the first set of flower sample images carries a flower sample label; The first flower sample image set is input into a pre-trained first image segmentation model to obtain the flower sample segmentation region output by the pre-trained first image segmentation model; the pre-trained first image segmentation model is obtained by training the first image segmentation model on a general image dataset; Based on the first loss value between the flower sample segmentation region and the flower sample label, the model parameters of the pre-trained first image segmentation model are fine-tuned until the first loss value meets the preset conditions, thus obtaining the first image segmentation model that has been trained.

[0010] According to a method for detecting bumblebee pollination effect provided by the present invention, before acquiring the image to be detected, the method further includes: Obtain a second set of flower sample images; the second set of flower sample images contains pollination marker sample labels; The second flower sample image set is input into the pre-trained second image detection model to obtain the pollination marker region detection box output by the second image detection model; Based on the second loss value between the pollination marker region detection box and the pollination marker sample label, the model parameters of the pre-trained second image detection model are fine-tuned until the second loss value meets the preset conditions, thus obtaining the trained second image detection model.

[0011] According to the present invention, a method for detecting bumblebee pollination effect is provided, wherein the image to be detected is a three-dimensional image, and the step of acquiring the image to be detected includes: Obtain two-dimensional images of the same crop from multiple angles; The three-dimensional image is obtained by performing three-dimensional reconstruction using the two-dimensional images from multiple angles.

[0012] According to the present invention, a method for detecting the pollination effect of bumblebees is provided, wherein the crop is blueberry.

[0013] The present invention also provides a bumblebee pollination effect detection device, comprising the following modules.

[0014] An image acquisition module is used to acquire an image to be detected; the image to be detected includes the flowers of a crop; The pollination mark detection module is used to input the detection image into the bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model can perform image recognition for each flower in the image to be detected, and obtain the pollination mark of each flower output by the bumblebee pollination mark recognition model. The pollination success rate calculation module is used to calculate the pollination success rate based on the number of successful pollination markers in the pollination markers and the total number of flowers. The bumblebee pollination marker recognition model that has been trained is based on flower sample images of the crop and pollination marker sample labels of each flower in the flower sample images of the crop.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the bumblebee pollination effect detection method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the bumblebee pollination effect detection method as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the bumblebee pollination effect detection method as described above.

[0018] This invention provides a method, apparatus, equipment, medium, and product for detecting bumblebee pollination effectiveness. The method involves acquiring an image to be detected, which includes crop flowers. The image is then input into a trained bumblebee pollination marker recognition model, which identifies each flower in the image and outputs a pollination marker for each flower. The pollination success rate is calculated based on the number of pollination markers and the total number of flowers. This method utilizes bumblebee pollination markers to automate the detection process of crop pollination success rate, improving the efficiency of bumblebee pollination effectiveness assessment and avoiding the time-consuming, labor-intensive, and subjective biases of manual observation. Furthermore, based on image analysis, pollination can be assessed without damaging the flowers, achieving non-destructive testing and providing rapid feedback on pollination effectiveness. This helps farmers promptly identify pollination problems (such as insufficient bee colony vitality or unsuitable environment) and take appropriate measures. Therefore, this method is flexible, simple, efficient, saves human resources, reduces costs, and can be widely adopted. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the bumblebee pollination effect detection method provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the bite holes on a blueberry flower provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the development process of a flower that is effectively pollinated by bumblebees biting holes, as provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the development process of a flower that has been pollinated ineffectively without bumblebee burrowing, as provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the bumblebee pollination effect detection device provided by the present invention.

[0025] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined Figures 1-6 Specific embodiments of the present invention are described below.

[0028] Figure 1 This is a flowchart illustrating the bumblebee pollination effect detection method provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps.

[0029] Step 101: Obtain the image to be detected; the image to be detected includes the flowers of the crop.

[0030] The image to be detected is a flower image of a specific crop, such as the flower of a tomato or blueberry. In particular, the accompanying drawings in this application use blueberry flowers as an example to illustrate the specific content of the scheme.

[0031] Specifically, high-resolution digital images of the flowers (especially the petals) to be tested are collected during the flowering period of blueberry crops. The image acquisition process can use fixed industrial cameras, handheld device cameras, or cameras mounted on drones to ensure that the blueberry petals are clearly visible and that the lighting conditions are good.

[0032] It should be noted that this application allows for the acquisition of flower images and the assessment of pollination success rates at different stages of blueberry flowering. For example, an assessment can be conducted once at the beginning of flowering or once during the peak flowering period.

[0033] Optionally, since bumblebees may leave bite holes (i.e., bumblebee pollination marks) at different locations on a flower, to improve the accuracy of bite hole identification and avoid missed or duplicate detections, this embodiment of the application can use a three-dimensional image to be detected. That is, a camera is used to acquire two-dimensional images of the same crop from different angles beforehand, and then these two-dimensional images from different angles are reconstructed in three dimensions to obtain a three-dimensional image of the flower. Specifically, step 101 above includes: acquiring two-dimensional images of the same crop from multiple angles; and using the two-dimensional images from multiple angles to perform three-dimensional reconstruction to obtain the three-dimensional image.

[0034] Step 102: Input the detected image into the bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model can perform image recognition for each flower in the image to be detected, and obtain the pollination mark of each flower output by the bumblebee pollination mark recognition model.

[0035] The trained bumblebee pollination mark recognition model refers to an artificial intelligence model used to identify and label bumblebee pollination marks in images. Bumblebee pollination marks are distinctive imprints left on flowers after a bumblebee pollinates them; different crops have different marks. For example, when a bumblebee visits a tomato flower, it uses its forehead to anchor itself to the flower, leaving a brown mark on the stigma called a "nose mark," which is the primary marker for identifying bumblebee pollination. When bumblebees pollinate blueberry flowers, "bite holes" appear on the stigma of the blueberry flower, such as... Figure 2 As shown, Figure 2 The image shows a "bite hole" on a blueberry blossom. These are specific damage marks on blueberry blossoms caused by bumblebee pollination and can be used to assess pollination success rates.

[0036] The trained bumblebee pollination mark recognition model can be derived from existing image detection or image segmentation models. For example, image detection models could be the YOLO series, Faster R-CNN, etc.; image segmentation models could be the U-Net series, FCN (Fully Convolutional Networks) models. Among these, the YOLO series focuses on fast detection and localization, suitable for real-time or large-scale applications. For higher localization accuracy or precise contours of pollination marks (e.g., for analyzing mark area, morphology, or distinguishing different subtle types of pollination marks), instance segmentation models such as Mask R-CNN, Cascade R-CNN, SOLOv2, or higher-precision two-stage detectors can be selected. This application uses the YOLOv8 model as an example for detailed explanation.

[0037] Considering that bumblebee pollination markers are usually small targets, the model selection should focus on its detection performance for small targets, or adopt techniques optimized for small target detection (such as improvements to the Feature Pyramid Network (FPN), the use of smaller and denser anchor boxes, attention mechanisms, etc.).

[0038] Specifically, the YOLOv8 model used in this embodiment is a single-stage object detection model that directly predicts the object category and bounding box in an end-to-end manner, and mainly consists of the following four parts.

[0039] 1. Input layer.

[0040] Receive the preprocessed image to be detected, typically an RGB three-channel image. The input size is adjusted according to the specific requirements of the selected model (e.g., 640x640, 800x800, or higher resolution to capture fine markings).

[0041] 2. Backbone network.

[0042] The backbone network is primarily used for feature extraction, also known as the feature extraction network, and is responsible for extracting multi-level deep feature maps from the input image. ResNet series, EfficientNet series, CSPNet series, Swin Transformer, ConvNeXt, etc., can be selected as the backbone network. For large-scale visual models, a larger capacity backbone network is usually used, or its powerful feature extraction capabilities obtained by pre-training on large general-purpose image datasets (such as ImageNet, COCO) are utilized.

[0043] 3. Neck network.

[0044] The neck network is used for feature fusion, which can effectively fuse feature maps from different levels of the backbone network, enhancing the detection capability of pollination markers at different scales. Especially for markers with large size variations, the neck network can adopt network structures such as FPN (Feature Pyramid Networks), PANet (Path Aggregation Network), BiFPN (Bidirectional Feature Pyramid Network), and PAFPN (Path Aggregation Feature Pyramid Network).

[0045] 4. Head network.

[0046] The head network, or detection head, is responsible for the final classification (determining whether it is a pollination marker) and regression (locating the position and extent of the pollination marker) based on the fused feature map. Among these, For image detection models, the head network outputs the class probability (confidence of the "pollen mark") and coordinates of each detection box (i.e., pollination mark).

[0047] For image segmentation models, the head network also outputs a pixel-level mask for each pollination marker.

[0048] Specifically, the image to be detected, containing the flowers of the crop, is input into the pre-trained bumblebee pollination marker recognition model (i.e., the YOLOv8 model). The YOLOv8 model identifies bumblebee bite holes on each flower in the image and encloses the identified objects in bounding boxes. Each bounding box carries a marker, such as "pollination marker," with 65% indicating a 65% confidence level that the marker is a bumblebee pollination marker. Marks greater than a preset threshold are used as the final pollination markers, indicating that the blueberry flower has been successfully pollinated.

[0049] Optionally, since a flower may have multiple pollination markers, it is also necessary to count the pollination markers on a flower. A flower is considered successfully pollinated if it contains at least one pollination marker. You can ultimately output only "Pollination Successful Marker". Alternatively, you can display "Pollination Successful" or "No Pollination Marker Detected" below each image.

[0050] Step 103: Calculate the pollination success rate based on the number of pollination markers and the total number of flowers.

[0051] Specifically, the ratio of the number of pollination markers to the total number of flowers is calculated, and this ratio is used as the pollination success rate.

[0052] Optionally, the pollination success rate can be calculated based on the ratio of the number of "successful pollination markers" to the total number of flowers. For example, if 85 pollination markers are found among 100 flowers, the pollination success rate is 85%.

[0053] It should be noted that the bumblebee pollination tag recognition model that has been trained above is based on the flower sample images of the crop and the pollination tag sample labels of each flower in the flower sample images of the crop.

[0054] Specifically, during the training process, sample images of crop flowers are first collected (at least 3,000-5,000 images of blueberry petals, or 3D images of blueberry flowers). These flower sample images contain pollination marker sample tags. These "pollen marker sample tags" are genuine tags that have been confirmed as bumblebee pollination markers.

[0055] Optionally, the flower sample image set used as training images may also include a large number of images identified as bumblebee pollination markers (i.e., positive samples), images of healthy petals where no such markers were observed (negative samples), and images with other non-pollination-induced damage (interference samples). These labels can be used by agricultural technicians or trained labelers to create bounding boxes for the pollination markers in the positive samples using tools such as LabelImg.

[0056] The above flower sample image set was preprocessed, such as image enhancement operations including random rotation (±20°), horizontal / vertical flipping, random cropping, brightness / contrast / saturation adjustment (±25%), Mosaic, MixUp, etc., to obtain more training samples.

[0057] Load COCO pre-trained weights, use the AdamW optimizer, with an initial learning rate of 0.001, and employ cosine annealing learning rate scheduling. Train for 150 epochs (training cycles) with a batch size of 32.

[0058] The preprocessed blueberry flower sample images are input into the YOLOv8 model, which outputs the bounding box, category (i.e., "pollen marker"), and confidence score of the pollination marker. The loss function typically consists of classification loss (e.g., cross-entropy loss, Focal Loss) and regression loss (e.g., Smooth L1 loss, GIoU / DIoU / CIoU loss, the latter being more robust to small objects and bounding box regression). The confidence threshold can be set to 0.5; markers with a confidence score greater than this threshold are considered final pollination markers. If at least one pollination marker with a confidence score greater than the threshold is detected on a single blueberry flower, the flower is considered successfully pollinated.

[0059] Optionally, if it is an image segmentation model, the loss function can consist of mask loss (such as binary cross-entropy loss, Dice Loss).

[0060] The above embodiment involves acquiring a target image, which includes crop flowers; inputting the target image into a trained bumblebee pollination marker recognition model, enabling the model to perform image recognition on each flower in the target image, and obtaining a pollination marker for each flower output by the model; calculating the pollination success rate based on the number of pollination markers and the total number of flowers. This method utilizes bumblebee pollination markers to automate the detection process of crop pollination success rate, improving the efficiency of bumblebee pollination effectiveness assessment, avoiding the time-consuming, labor-intensive, and subjective biases of manual observation; and based on image analysis, it assesses pollination without damaging the flowers, achieving non-destructive testing and providing rapid feedback on pollination effectiveness, helping farmers to promptly identify pollination problems (such as insufficient bee colony vitality, unsuitable environment for pollination, etc.) and take corresponding measures. Therefore, this method is flexible, simple, efficient, saves human resources, reduces costs, and can be widely promoted.

[0061] In one embodiment, the trained bumblebee pollination marker recognition model includes a trained first image segmentation model and a trained second image detection model, and step 102 includes the following steps.

[0062] The detected image is input into the first image segmentation model that has been trained, and a flower segmentation image containing flower region markers is obtained from the output of the first image segmentation model; The flower segmentation image is input into the trained second image detection model to obtain the authorization label for the flower in the flower region label output by the second image detection model.

[0063] The first image segmentation model, which has been trained, is an image segmentation model built on the U-Net model. It is used to segment the image to be detected and obtain a mask image of the flower / petal. The second image detection model, which has been trained, is an image detection model built on the YOLOv8 model. It is used to further segment the flower / petal image obtained above to obtain a detection box for the bite hole, i.e., to obtain the pollination mark.

[0064] Specifically, the images to be detected are first preprocessed, including image enhancement and Gaussian denoising. The lightweight U-Net model is then used to extract flower features (e.g., color and shape features) from the images, automatically extracting regions of single or multiple blueberry flowers, and further focusing on the corolla tube (petals). The extracted petal / flower region images are scaled to the YOLOv8 model input size (e.g., 640x640 pixels) and normalized. The normalized image is then input into the trained YOLOv8 model to obtain the detection bounding box output by the YOLOv8 model. This bounding box contains the bite hole and its confidence level.

[0065] The above embodiment combines an image segmentation model and an image detection model to detect pollination markers in two stages. First, the flower / petal image is segmented using an image segmentation method to filter out interference from other regions, such as branches and leaves. Then, the puncture holes are detected from the flower / petal image, which can improve the detection accuracy of pollination markers.

[0066] In one embodiment, the above step 101 is preceded by the following steps.

[0067] A first flower sample image set is obtained; each flower sample image in the first flower sample image set carries a flower sample label; the first flower sample image set is input into a pre-trained first image segmentation model to obtain the flower sample segmentation region output by the pre-trained first image segmentation model; the pre-trained first image segmentation model is obtained by training the first image segmentation model on a general image dataset; based on the first loss value between the flower sample segmentation region and the flower sample label, the model parameters of the pre-trained first image segmentation model are fine-tuned until the first loss value meets a preset condition, thus obtaining the first image segmentation model that has been trained.

[0068] The first flower sample image set consists of images labeled with blueberry flower / petal sample tags, which were pre-labeled by agricultural technicians. The images in the first flower sample image set should cover flowers / petals at different flowering periods, under different lighting conditions, at different shooting angles, and with different shapes and resolutions.

[0069] It should be noted that the model training in this embodiment adopts the transfer learning method, that is, the model weights pre-trained on a large general image dataset are used as the initial weights, and then fine-tuning is performed on a specially constructed image set containing the above-mentioned first flower sample image set, so as to finally obtain the first image segmentation model after training.

[0070] Specifically, the aforementioned first flower sample image set is input into a pre-trained first image segmentation model (e.g., a U-Net model). This pre-trained first image segmentation model can be a model pre-trained on a large general-purpose image dataset (such as ImageNet or COCO), thus enabling the model to acquire powerful feature extraction capabilities in advance. After the aforementioned first flower sample image set is segmented by the pre-trained first image segmentation model, flower (or petal) sample segmentation regions are obtained. A first loss function (e.g., a binary cross-entropy loss function) is used to calculate the first loss value between the aforementioned flower (or petal) sample segmentation regions and the flower sample labels. The model parameters of the aforementioned pre-trained first image segmentation model are then fine-tuned until the aforementioned first loss value meets a preset condition, resulting in a trained first image segmentation model.

[0071] The above embodiments first pre-train the model on a large general image dataset using the transfer learning method, use the pre-trained model weights as initial weights, and then fine-tune the model parameters on a specially constructed flower / petal label dataset, which can improve the model's generalization ability.

[0072] Furthermore, the above embodiments, through learning from a large number of real pollination marker samples using an artificial intelligence model, can accurately identify pollination markers with diverse morphologies, even those that are relatively hidden, outperforming traditional image processing methods. In addition, the accumulated pollination marker data and success rate data can be used to analyze key factors affecting pollination efficiency, guiding precision agriculture practices.

[0073] In one embodiment, before step 101, the method further includes: acquiring a second flower sample image set; the second flower sample image set containing pollination marker sample labels; inputting the second flower sample image set into a pre-trained second image detection model to obtain a pollination marker region detection box output by the second image detection model; and fine-tuning the model parameters of the pre-trained second image detection model based on a second loss value between the pollination marker region detection box and the pollination marker sample labels until the second loss value meets a preset condition, thereby obtaining the trained second image detection model.

[0074] Similarly, the second flower sample image set can be flower / petal sample images containing pollination marker sample labels. Furthermore, these images can be augmented, for example, by extensively using geometric transformations (random rotation, flipping, scaling, translation, cropping – ensuring the integrity of the markers), color and lighting transformations (brightness, contrast, saturation, hue adjustment, Gaussian blur, noise injection), and advanced data augmentation strategies (MixUp, CutMix, Mosaic) to obtain more sample images, which helps improve the model's generalization ability. The aforementioned second image detection model can be a YOLOv8 model. Again, the trained second image detection model is obtained using transfer learning. The second loss value is obtained based on a second loss function, which can be, for example, cross-entropy loss or Focal Loss.

[0075] It should be noted that the second image detection model can not only output pollination marker region detection boxes, that is, mark the identified bumblebee pollination markers with bounding boxes or highlighted areas, but also clearly indicate whether the flower has been determined to be "successfully pollinated," and can also output the location information of the pollination marker in the image.

[0076] The above embodiments first pre-train the model on a large general image dataset using the transfer learning method, use the pre-trained model weights as initial weights, and then fine-tune the model parameters on a specially constructed pollination marker dataset, which can improve the model's generalization ability.

[0077] Furthermore, the experimental data upon which the method proposed in this application is based are as follows.

[0078] 1. Experimental Materials The test bee colony came from ground bumblebees raised at the bumblebee breeding base. The blueberry variety was three-year-old Eurica. The greenhouse area was 1,000 square meters. When the flowers were 5% open, a colony of about 60 ground bumblebees was placed in a box.

[0079] 2. Experimental Methods More than 100 flower buds were randomly marked. Each marked flower was observed and photographed daily, and the flowering time, the time of being bitten, the time of flower fall, and the time of fruit formation were recorded. The experiment was repeated three times. The development process of flowers effectively pollinated by bumblebees through their holes is as follows: Figure 3 As shown, the development process of a flower that has been ineffectively pollinated without bumblebee burrowing is as follows: Figure 4 As shown.

[0080] 3. Experimental Results After three repeated observations of the flowers marked with bumblebee holes and the results of fruit setting, it was found that the fruit setting rate of flowers with holes was 85%, while the fruit setting rate of flowers without holes quickly withered and fell off, with a fruit setting rate of less than 5%. This further verifies that the hole marking can quickly determine whether blueberry flowers are sufficiently pollinated, allowing for timely evaluation of the bumblebee pollination effect. Based on the evaluation results, the pollination plan can be adjusted and optimized, and the bee colonies can be adjusted or replaced in a timely manner, thereby saving time and achieving efficient and precise pollination and reasonable allocation of bumblebees.

[0081] The bumblebee pollination effect detection device provided by the present invention is described below. The bumblebee pollination effect detection device described below can be referred to in correspondence with the bumblebee pollination effect detection method described above.

[0082] like Figure 5 As shown, Figure 5 A schematic diagram of a bumblebee pollination effect detection device is shown, which includes the following modules.

[0083] Image acquisition module 501 is used to acquire an image to be detected; the image to be detected includes the flowers of a crop; The pollination mark detection module 502 is used to input the detection image into the bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model can perform image recognition for each flower in the image to be detected, and obtain the pollination mark of each flower output by the bumblebee pollination mark recognition model. The pollination success rate calculation module 503 is used to calculate the pollination success rate based on the number of successful pollination markers in the pollination markers and the total number of flowers. The bumblebee pollination marker recognition model that has been trained is based on flower sample images of the crop and pollination marker sample labels of each flower in the flower sample images of the crop.

[0084] In one embodiment, the trained bumblebee pollination marker recognition model includes a trained first image segmentation model and a trained second image detection model; the pollination marker detection module 502 is further configured to: The detected image is input into the first image segmentation model that has been trained, and a flower segmentation image containing flower region markers is obtained from the output of the first image segmentation model; The flower segmentation image is input into the trained second image detection model to obtain the authorization label for the flower in the flower region label output by the second image detection model.

[0085] In one embodiment, the above-described apparatus further includes a model training unit, which is used for: Obtain a first set of flower sample images; each flower sample image in the first set of flower sample images carries a flower sample label; The first flower sample image set is input into a pre-trained first image segmentation model to obtain the flower sample segmentation region output by the pre-trained first image segmentation model; the pre-trained first image segmentation model is obtained by training the first image segmentation model on a general image dataset; Based on the first loss value between the flower sample segmentation region and the flower sample label, the model parameters of the pre-trained first image segmentation model are fine-tuned until the first loss value meets the preset conditions, thus obtaining the first image segmentation model that has been trained.

[0086] In one embodiment, the model training unit is further configured to: Obtain a second set of flower sample images; the second set of flower sample images contains pollination marker sample labels; The second flower sample image set is input into the pre-trained second image detection model to obtain the pollination marker region detection box output by the second image detection model; Based on the second loss value between the pollination marker region detection box and the pollination marker sample label, the model parameters of the pre-trained second image detection model are fine-tuned until the second loss value meets the preset conditions, thus obtaining the trained second image detection model.

[0087] In one embodiment, the image to be detected is a three-dimensional image, and the image acquisition module 601 is further configured to: Obtain two-dimensional images of the same crop from multiple angles; The three-dimensional image is obtained by performing three-dimensional reconstruction using the two-dimensional images from multiple angles.

[0088] In one embodiment, the crop is blueberry.

[0089] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call logical instructions in the memory 630 to execute a bumblebee pollination effect detection method, which includes: acquiring an image to be detected; the image to be detected includes crop flowers; inputting the image to be detected into a bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model performs image recognition on each flower in the image to be detected, and obtains the pollination mark of each flower output by the bumblebee pollination mark recognition model; calculating the pollination success rate based on the number of pollination marks and the total number of flowers; wherein the bumblebee pollination mark recognition model that has been trained is based on sample images of crop flowers and pollination mark sample labels in the sample images of crop flowers.

[0090] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0091] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the bumblebee pollination effect detection method provided by the above methods. The method includes: acquiring an image to be detected; the image to be detected includes flowers of a crop; inputting the image to be detected into a bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model performs image recognition for each flower in the image to be detected, and obtains a pollination mark for each flower output by the bumblebee pollination mark recognition model; calculating the pollination success rate based on the number of pollination marks and the total number of flowers; wherein the bumblebee pollination mark recognition model that has been trained is based on sample images of flowers of the crop and pollination mark sample labels in the sample images of flowers of the crop.

[0092] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the bumblebee pollination effect detection method provided by the above methods. The method includes: acquiring an image to be detected; the image to be detected includes flowers of a crop; inputting the detected image into a bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model performs image recognition for each flower in the image to be detected, and obtains a pollination mark for each flower output by the bumblebee pollination mark recognition model; calculating the pollination success rate based on the number of pollination marks and the total number of flowers; wherein the bumblebee pollination mark recognition model that has been trained is based on sample images of flowers of the crop and pollination mark sample labels in the sample images of flowers of the crop.

[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0094] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the pollination effect of bumblebees, characterized in that, include: Acquire an image to be detected; the image to be detected includes the flowers of a crop; The detected image is input into a bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model can perform image recognition for each flower in the image to be detected, and obtain the pollination mark of each flower output by the bumblebee pollination mark recognition model. The pollination success rate is calculated based on the number of pollination markers and the total number of flowers. The bumblebee pollination marker recognition model that has been trained is based on flower sample images of the crop and pollination marker sample labels in the flower sample images of the crop.

2. The method for detecting bumblebee pollination effect according to claim 1, characterized in that, The trained bumblebee pollination mark recognition model includes a trained first image segmentation model and a trained second image detection model. The detected image is input into the trained bumblebee pollination mark recognition model, so that the model performs image recognition on each flower in the image to be detected, obtaining the pollination mark of each flower output by the model, including: The detected image is input into the first image segmentation model that has been trained, and a flower segmentation image containing flower region markers is obtained from the output of the first image segmentation model; The flower segmentation image is input into the trained second image detection model to obtain the authorization label for the flower in the flower region label output by the second image detection model.

3. The method for detecting bumblebee pollination effect according to claim 2, characterized in that, Before acquiring the image to be detected, the method further includes: Obtain a first set of flower sample images; each flower sample image in the first set of flower sample images carries a flower sample label; The first flower sample image set is input into a pre-trained first image segmentation model to obtain the flower sample segmentation region output by the pre-trained first image segmentation model; the pre-trained first image segmentation model is obtained by training the first image segmentation model based on a general image dataset; Based on the first loss value between the flower sample segmentation region and the flower sample label, the model parameters of the pre-trained first image segmentation model are fine-tuned until the first loss value meets the preset conditions, thus obtaining the first image segmentation model that has been trained.

4. The method for detecting bumblebee pollination effect according to claim 3, characterized in that, Before acquiring the image to be detected, the method further includes: Obtain a second set of flower sample images; the second set of flower sample images contains pollination marker sample labels; The second flower sample image set is input into the pre-trained second image detection model to obtain the pollination marker region detection box output by the second image detection model; Based on the second loss value between the pollination marker region detection box and the pollination marker sample label, the model parameters of the pre-trained second image detection model are fine-tuned until the second loss value meets the preset conditions, thus obtaining the trained second image detection model.

5. The method for detecting bumblebee pollination effect according to claim 1, characterized in that, The image to be detected is a three-dimensional image, and the acquisition of the image to be detected includes: Obtain two-dimensional images of the same crop from multiple angles; The three-dimensional image is obtained by performing three-dimensional reconstruction using the two-dimensional images from multiple angles.

6. The method for detecting bumblebee pollination effectiveness according to any one of claims 1 to 5, characterized in that, The crop in question is blueberry.

7. A device for detecting the pollination effect of bumblebees, characterized in that, include: An image acquisition module is used to acquire an image to be detected; the image to be detected includes the flowers of a crop; The pollination mark detection module is used to input the detection image into the bumblebee pollination mark recognition model that has been trained, so that the bumblebee pollination mark recognition model can perform image recognition for each flower in the image to be detected, and obtain the pollination mark of each flower output by the bumblebee pollination mark recognition model. The pollination success rate calculation module is used to calculate the pollination success rate based on the number of successful pollination markers in the pollination markers and the total number of flowers. The bumblebee pollination marker recognition model that has been trained is based on flower sample images of the crop and pollination marker sample labels of each flower in the flower sample images of the crop.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the bumblebee pollination effect detection method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the bumblebee pollination effect detection method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the bumblebee pollination effect detection method as described in any one of claims 1 to 6.

Images