Corn emergence rate and emergence uniformity evaluation method, device, terminal and medium

Abstract

The application discloses a corn emergence rate and emergence uniformity evaluation method, device, terminal and medium, the method determines a plurality of corn plant images according to a remote sensing image of a target area; an improved YOLOv10 model is used to predict the corresponding rotating bounding box and key point of each corn plant image, the key point is the center of the corn plant; the corn emergence rate and the emergence uniformity corresponding to the target area are determined according to the rotating bounding box and the key point corresponding to each corn plant image. The improved YOLOv10 model is used to predict the rotating bounding box and the key point of the corn plant, the rotating bounding box can effectively reduce the interference of the background or other non-crop areas, improve the detection accuracy, and the center of the corn plant is accurately marked through the key point, so that the distance and distribution between plants can be accurately evaluated, and the problem that the center of the corn plant is marked by the center point of the detected target frame in the prior art, resulting in a large detection error.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of remote sensing, and particularly relates to a corn emergence rate and emergence uniformity evaluation method and device, a terminal and a medium. BACKGROUND

[0002] Corn is one of the main food crops widely planted around the world, and its production is related to food security and economic benefits. Evaluating the emergence rate and uniformity of corn can help farmers and agricultural technicians quickly identify problems in the planting process, such as improper seeding depth, poor soil conditions, and low seed quality, so that remedial measures can be taken to ensure higher crop yields and more rational resource utilization.

[0003] By mounting a camera on a drone to take high-resolution corn images from the air, agronomic researchers can quickly obtain corn information for large areas of farmland. Currently, the processing of unmanned aerial vehicle remote sensing data usually uses image recognition algorithms, such as using a target detection method to automatically label corn, and framing each corn in a target box, and marking the center of the corn through the center point of the detected target box. However, the actual center of the corn is often not at the geometric center of the target box, especially in the case of uneven emergence or plant shading, resulting in large detection errors.

[0004] Therefore, the prior art still needs to be improved and developed. SUMMARY

[0005] The technical problem solved by the present application is to provide a corn emergence rate and emergence uniformity evaluation method, device, terminal and medium to solve the problem of large detection errors caused by marking the center of the corn plant through the center point of the detected target box in the prior art.

[0006] The technical solution adopted by the present application to solve the problem is as follows:

[0007] In a first aspect, the embodiments of the present application provide a corn emergence rate and emergence uniformity evaluation method, wherein the method comprises:

[0008] obtaining a remote sensing image of a target area, and determining a plurality of corn plant images according to the remote sensing image;

[0009] using an improved YOLOv10 model to predict according to the plurality of corn plant images, and determining a rotation bounding box and a key point corresponding to each of the corn plant images, wherein the improved YOLOv10 model is realized by introducing a Mamba attention module in the neck network layer of the YOLOv10 model, and the key point is used to label the center of the corn plant;

[0010] Determine the corn emergence rate and emergence uniformity of the target region according to the rotation bounding box and the key point corresponding to each of the corn plant images.

[0011] In an implementation method, the determining the plurality of corn plant images according to the remote sensing image comprises:

[0012] Extract the inter-row straight line of the corn plant in the remote sensing image, and determine the corn plant row in the remote sensing image according to the inter-row straight line.

[0013] According to the corn plant row, the remote sensing image is segmented, and a plurality of corn plant images are determined.

[0014] In an implementation method, the extracting the inter-row straight line of the corn plant in the remote sensing image comprises:

[0015] Perform image enhancement on the corn plant in the remote sensing image, and convert the enhanced remote sensing image into a binary image.

[0016] Perform morphological operation and Hough transform on the binary image, and extract the inter-row straight line of the corn plant in the remote sensing image.

[0017] In an implementation method, the training method of the improved YOLOv10 model comprises:

[0018] Obtain a plurality of historical remote sensing images, and perform rotation bounding box and key point labeling on each of the historical remote sensing images, and use the labeled plurality of historical remote sensing images as training data.

[0019] Construct a loss function corresponding to the improved YOLOv10 model.

[0020] Train the improved YOLOv10 model according to the training data and the loss function.

[0021] In an implementation method, the constructing a loss function corresponding to the improved YOLOv10 model comprises:

[0022] Calculate the classification loss, bounding box loss and key point loss in the training process of the improved YOLOv10 model.

[0023] Weight the classification loss, the bounding box loss and the key point loss to determine the loss function corresponding to the improved YOLOv10 model.

[0024] In an implementation method, the determining the corn emergence rate of the target region according to the rotation bounding box corresponding to each of the corn plant images comprises:

[0025] determine a number of rotation boxes corresponding to the target region according to the rotation bounding boxes corresponding to each of the corn plant images;

[0026] obtain an area of the target region, and determine the corn emergence rate of the target region according to the number of rotation boxes and the area of the target region.

[0027] In an implementation method, the uniformity of emergence of the target region is determined according to the key points corresponding to each of the corn plant images, including:

[0028] determine a pixel distance between each pair of adjacent corn plants according to the key points corresponding to each of the corn plant images;

[0029] obtain a ground sampling distance, and determine an actual distance between each pair of adjacent corn plants according to the pixel distance and the ground sampling distance;

[0030] determine the uniformity of emergence of the target region according to the actual distance between each pair of adjacent corn plants.

[0031] In a second aspect, an embodiment of the present application further provides a corn emergence rate and uniformity of emergence evaluation device, wherein the corn emergence rate and uniformity of emergence evaluation device comprises:

[0032] a collection module configured to obtain a remote sensing image of a target region, and determine a plurality of corn plant images according to the remote sensing image;

[0033] a prediction module configured to predict according to the plurality of corn plant images by using an improved YOLOv10 model, and determine a rotation bounding box and a key point corresponding to each of the corn plant images, wherein the improved YOLOv10 model is realized by introducing a Mamba attention module in a neck network layer of the YOLOv10 model, and the key point is used to mark a center of a corn plant;

[0034] an evaluation module configured to determine a corn emergence rate and a uniformity of emergence of the target region according to the rotation bounding box and the key point corresponding to each of the corn plant images.

[0035] In a third aspect, an embodiment of the present application further provides a terminal, which comprises a memory and one or more processors; the memory stores one or more programs; the programs contain instructions for executing the corn emergence rate and uniformity of emergence evaluation method of any of the above; and the processor is configured to execute the programs.

[0036] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a plurality of instructions, wherein the instructions are adapted to be loaded and executed by a processor to implement the corn emergence rate and uniformity of emergence evaluation method of any of the above.

[0037] The application has the beneficial effects that: the embodiment of the application determines a plurality of corn plant images according to a remote sensing image of a target area; adopts an improved YOLOv10 model to predict a rotation bounding box and a key point corresponding to each corn plant image, the key point being the center of the corn plant; and determines the corn emergence rate and the emergence uniformity corresponding to the target area according to the rotation bounding box and the key point corresponding to each corn plant image. Since the improved YOLOv10 model is adopted to predict the rotation bounding box and the key point of the corn plant, the rotation bounding box can effectively reduce the interference of the background or other non-crop areas, improve the detection accuracy, and the key point can accurately mark the center of the corn plant, so that the distance and distribution between plants can be more accurately evaluated, thus effectively solving the problem that the center of the corn plant is marked by the center point of the detected target frame in the prior art, resulting in a large detection error. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments described in the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0039] Figure 1 is a flowchart of the corn emergence rate and emergence uniformity evaluation method provided by the embodiment of the application.

[0040] Figure 2 is an embodiment flowchart of the corn emergence rate and emergence uniformity evaluation method provided by the embodiment of the application.

[0041] Figure 3 is an improved YOLOv10 model architecture diagram provided by the embodiment of the application.

[0042] Figure 4 is a Mamba attention module architecture diagram provided by the embodiment of the application.

[0043] Figure 5 is a rotation bounding box and key point prediction diagram of a small corn plant provided by the embodiment of the application.

[0044] Figure 6 is a rotation bounding box and key point prediction diagram of a large corn plant provided by the embodiment of the application.

[0045] Figure 7 is a rotation bounding box and key point prediction diagram of a corn plant containing interference plants provided by the embodiment of the application.

[0046] Figure 8 is a schematic diagram of an internal module of the corn emergence rate and emergence uniformity evaluation device provided by the embodiment of the present application.

[0047] Figure 9 is a principle block diagram of the terminal provided by the embodiment of the present application. DETAILED DESCRIPTION

[0048] The present application discloses a corn emergence rate and emergence uniformity evaluation method, device, terminal and medium. In order to make the purpose, technical scheme and effect of the present application more clear and explicit, the present application will be further described in detail below with reference to the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0049] Those skilled in the art can understand that, unless specifically stated otherwise, the singular forms "a", "an" and "the" used herein also include the plural forms. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" used herein can include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any single unit and all combinations of the associated listed items.

[0050] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as that generally understood by those skilled in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have meanings consistent with those in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as such.

[0051] Corn is one of the main food crops widely planted around the world, and its production is related to food security and economic benefits. Evaluating the emergence rate and uniformity of corn can help farmers and agricultural technicians quickly find problems in the planting process, such as improper seeding depth, poor soil conditions, low seed quality, etc., so as to take remedial measures to ensure higher crop yield and more rational resource utilization.

[0052] By taking high-resolution corn images from the air with a camera mounted on a drone, agronomists can quickly obtain corn information for large areas of farmland. Currently, the processing of unmanned aerial vehicle remote sensing data usually uses image recognition algorithms, such as using a target detection method to automatically label corn, and framing each corn in a target box, and marking the center of the corn through the center point of the detected target box. However, the actual center of the corn is often not at the geometric center of the target box, especially in the case of uneven emergence or plant shading, resulting in large detection errors.

[0053] In view of the above defects of the prior art, the present application provides a corn emergence rate and emergence uniformity evaluation method, which determines a plurality of corn plant images according to a remote sensing image of a target area; uses an improved YOLOv10 model to predict the rotation bounding box and key points corresponding to each corn plant image, the key points being the center of the corn plant; determines the corn emergence rate and the emergence uniformity corresponding to the target area according to the rotation bounding box and the key points corresponding to each corn plant image. Since the improved YOLOv10 model is used to predict the rotation bounding box and the key points of the corn plant in the present application, the rotation bounding box can effectively reduce the interference of the background or other non-crop areas, improve the detection accuracy, and the key points can accurately mark the center of the corn plant, so that the distance and distribution between plants can be more accurately evaluated, thereby effectively solving the problem that the center of the corn plant is marked by the center point of the detected target box in the prior art, resulting in large detection errors.

[0054] Exemplary method

[0055] As shown in Figure 1 , the method comprises:

[0056] Step S100, obtaining a remote sensing image of a target area, and determining a plurality of corn plant images according to the remote sensing image.

[0057] In short, the present embodiment uses a drone to carry an RGB camera to obtain a high-resolution remote sensing image of a target area. Compared with satellite optical and radar remote sensing technology, the use of a drone to carry an RGB camera to collect remote sensing images has the characteristics of high mobility and high precision, which is beneficial to timely information acquisition and image interpretation. The drone flies at low altitude to capture clear image data, ensuring that the state of each corn seedling can be effectively recorded. The use of a drone to carry an RGB camera for image acquisition can cover a large area of farmland in a short time, with high efficiency and good precision, and is suitable for large-scale farmland monitoring. In addition to drones, ground mobile robots or low-altitude aircraft (such as balloons, gliders, etc.) can also be used for image acquisition according to different environmental conditions. The collected image data can be selected according to different environmental conditions, such as multispectral images, thermal images, etc.

[0058] After obtaining the remote sensing image of the target area, the corn plant image in the remote sensing image is extracted, so as to label the rotation bounding box and key points of each corn plant image.

[0059] In an implementation manner, the corn plant images are determined according to the remote sensing image, including:

[0060] In step S101, the inter-row straight line of the corn plant in the remote sensing image is extracted, and the corn plant row in the remote sensing image is determined according to the inter-row straight line.

[0061] In step S102, the remote sensing image is segmented according to the corn plant row, and a plurality of corn plant images are determined.

[0062] Specifically, corn is usually planted in a row unit. After the inter-row straight line of the corn is extracted, the planting row and column of the crop in the target area are quickly detected according to the inter-row straight line of the corn, the corn plant row is obtained, and the interference of irrelevant factors such as soil on subsequent detection is reduced.

[0063] According to the identified corn plant row, the original high-resolution remote sensing image is segmented, and each corn plant in the remote sensing image is ensured to be complete in the cropped image, as shown in Figure 2 .

[0064] In order to improve the accuracy of subsequent labeling of the rotation bounding box and key points of the corn plant, the embodiment also performs image enhancement on the segmented corn plant image to improve the quality and diversity of the corn plant image. For example, image preprocessing is performed under different weather and light conditions to reduce the influence of the environment on the detection result, multiple images are spliced to increase the diversity of the data, new images are generated through linear interpolation to enhance the learning of the model on the gradual change information, and the brightness, contrast, saturation and hue of the image are randomly adjusted to improve the inference ability of the model under different light conditions.

[0065] In an implementation manner, the inter-row straight line of the corn plant in the remote sensing image is extracted, including:

[0066] In step S1011, the corn plant in the remote sensing image is image enhanced, and the enhanced remote sensing image is converted into a binary image.

[0067] In step S1012, morphological operation and Hough transformation are performed on the binary image, and the inter-row straight line of the corn plant in the remote sensing image is extracted.

[0068] Specifically, the embodiment performs image enhancement on the corn plants in the remote sensing image through a green channel or a normalized vegetation index, highlighting the green area of the corn plants. The normalized vegetation index (NDVI index) is a remote sensing index reflecting the ground vegetation coverage. The calculation formula is: NDVI=(NIR-R) / (NIR+R), where NIR is the reflection value of the near-infrared band, and R is the reflection value of the red light band.

[0069] For the enhanced remote sensing image, the color or texture features (such as the green channel, the normalized vegetation index, etc.) of the vegetation are used to distinguish the crops from the background. The embodiment uses a threshold segmentation technique to generate a binary image from the remote sensing image, in which the crops are white and the background is black.

[0070] The binary image is smoothed by morphological operations (such as opening operation or closing operation) to remove noise and small areas and retain continuous corn plant areas. Then, the Hough transform is performed on the binary image after the morphological operation to extract the inter-row straight lines of the corn plants, so as to quickly detect the median row and column of the corn plants in the target area.

[0071] In step S200, an improved YOLOv10 model is used to predict the rotation bounding box and key points corresponding to each of the corn plant images according to the corn plant images, wherein the improved YOLOv10 model is realized by introducing a Mamba attention module into the neck network layer of the YOLOv10 model, and the key points are used to mark the center of the corn plant.

[0072] The YOLOv10 model has high-speed and high-precision detection capability. The model proposes a lightweight classification head, spatial-channel decoupled down-sampling, and block design guided by ordering to reduce obvious computational redundancy and realize a more efficient architecture. In addition, in order to improve the accuracy, the model explores large kernel convolution and proposes an effective partial self-attention (PSA) module to enhance the understanding ability of the model.

[0073] As shown in Figure 3 The embodiment introduces a Mamba attention module into the neck network layer of the YOLOv10 model to obtain an improved YOLOv10 model, and the Mamba attention module is as shown in Figure 4The Mamba attention module combines position embeddings, facilitating precise location-based visual understanding. By incorporating the Mamba module into the YOLOv10 model, which can be quickly built and optimized, the linear complexity of CNN and the global receptive field of the Mamba attention module are utilized to effectively detect the emergence of corn plants, not only improving the detection accuracy of the model in key areas, but also significantly optimizing the running speed of the model and reducing the memory occupation in the calculation process. The improved YOLOv10 model can maintain high precision while significantly improving the efficiency and resource utilization of the model, making it particularly suitable for large-scale real-time monitoring and evaluation tasks in farmland.

[0074] In particular, Mamba is a sequence modeling architecture that leverages a selective state space model (SSM) to efficiently handle long sequences. It addresses the computational inefficiency of Transformer architectures, especially when dealing with long sequences, by introducing a mechanism that allows the model to selectively propagate or forget information based on the current input. This is achieved by making SSM parameters a function of the input, thereby enhancing the model's ability to perform content-based reasoning. Mamba operates in two modes: a recurrent mode for autoregressive reasoning and a convolutional mode for parallel training. The architecture is simplified by integrating the selective SSM into a single block without the need for attention or multi-layer perceptron (MLP) components, resulting in a streamlined design that maintains a linear time complexity with respect to sequence length.

[0075] The Mamba attention module combines adaptive average pooling and Mamba modules to enhance the feature representation capabilities of deep learning models. The module takes an input tensor x ∈ R B×C×H×W , where B represents the batch size, C represents the number of channels, H and W represent the height and width, respectively. The module generates two feature representations x x and x y through adaptive average pooling along the width and height dimensions:

[0076] x x = AdaptiveAvgPool2d(x,(None,1)),

[0077] x y = AdaptiveAvgPool2d(x,(1,None)),

[0078] During this process, x x compresses the width of the input feature map while preserving the height information, and x y compresses the height while preserving the width information. The module utilizes Mamba to process the aforementioned features, extracting important information. For x x , the output representation is x ma :

[0079] x ma = Mamba(x x ),

[0080] The Sigmoid activation function is applied to introduce nonlinearity and generate dynamic feature weights:

[0081] x x = σ(x ma ),

[0082] x y = σ(Mamba(x y )),

[0083] x x and x y represent the generated weights in the width and height dimensions, allowing the model to dynamically adjust its attention to different regions of the input features. Finally, the Mamba attention module combines the input tensor with the processed weights through element-wise multiplication to obtain the output:

[0084]

[0085] where ⊙ denotes element-wise multiplication, represents the expansion of each channel. This process ensures that the model can flexibly focus on the key parts of the input features, significantly improving its ability to focus on important features and overall performance. In addition, this module aims to optimize the feature selection process, reducing computational complexity while maintaining high efficiency, in order to better segment mutually occluded weeds.

[0086] Since the growth direction of corn plants in the field may be inclined, this embodiment can more accurately adapt to the growth direction of corn crops by using the improved YOLOv10 model to predict a rotated bounding box instead of a traditional horizontal bounding box to frame the plant area. Not only can the angle and orientation of the corn plant be considered, but the boundaries of the plant can also be more accurately delineated. At the same time, by using the improved YOLOv10 model to predict key points, the center position of the corn plant can be accurately labeled, effectively improving the accuracy of the uniformity evaluation of the corn plant.

[0087] In order to enable the improved YOLOv10 model to accurately predict the rotated bounding box and key points corresponding to the corn plant image, the improved YOLOv10 model is first trained in the embodiment. The training method of the improved YOLOv10 model includes: obtaining a plurality of historical remote sensing images, performing rotated bounding box and key point labeling on each of the historical remote sensing images, taking the labeled historical remote sensing images as training data, and the plurality of historical remote sensing images contain each growth period of corn seedlings; constructing a loss function corresponding to the improved YOLOv10 model; training the improved YOLOv10 model according to the training data and the loss function to obtain a trained improved YOLOv10 model, so as to predict the growth period, the rotated bounding box and the key points of the corn plant according to the improved YOLOv10 model.

[0088] In an implementation manner, the constructing the loss function corresponding to the improved YOLOv10 model comprises:

[0089] calculating the classification loss, the bounding box loss and the key point loss in the training process of the improved YOLOv10 model;

[0090] weighting the classification loss, the bounding box loss and the key point loss to determine the loss function corresponding to the improved YOLOv10 model.

[0091] Since the rotated bounding box and key points corresponding to the corn plant image are predicted by the improved YOLOv10 model, and the classification of the corn plant and other elements is also needed, the classification loss, the bounding box loss and the key point loss in the training process of the improved YOLOv10 model are weighted in the embodiment to obtain the loss function corresponding to the improved YOLOv10 model.

[0092] 1. The classification loss (ClassLoss) is a binary cross entropy loss (BinaryCrossEntropy Loss, BCE), and the formula is as follows:

[0093]

[0094] y i is the class label of the corn plant, which is used to represent the growth period of the corn plant, p i is the probability predicted by the improved YOLOv10 model, and N is the total number of corn plants.

[0095] 2. The bounding box loss (BboxLoss) includes a rotated bounding box loss (RotatedBboxLoss) and a distribution focal loss (DistributionFocalLoss, DFL).

[0096] The formula of the rotated bounding box loss is as follows:

[0097] L bbox = IoU rotated (B pred , B gt ),

[0098] where B pred is the rotated bounding box predicted by the improved YOLOv10 model, B gt is the real rotated bounding box, and the IoU or GIoU is used for calculation.

[0099] The calculation formula of the distribution focus loss is as follows:

[0100] L dfl = Softmax (dist) · proj,

[0101] where disf is the distribution predicted by the improved YOLOv10 model, and proj is the projection operation used to calculate the parameters of the rotated bounding box in the discrete distribution.

[0102] 3. The key point loss (Keypoint Loss) includes a key point position loss (Keypoint position Loss) and a key point object loss (Keypoint Object Loss).

[0103] The calculation formula of the key point position loss is as follows:

[0104]

[0105] where is the key point coordinate predicted by the improved YOLOv10 model, u i is the real key point, and A i is the target area area, used to balance the importance of the key point.

[0106] The calculation formula of the key point object loss is as follows:

[0107]

[0108] where is the key point visibility label predicted by the improved YOLOv10 model, h i is the real label, and the binary cross entropy is used for calculation.

[0109] 4. The classification loss, the bounding box loss, and the key point loss are weighted to determine the loss function corresponding to the improved YOLOv10 model, which is represented as follows:

[0110] L total = Lbbox ·w box +L kpts ·w pose +L kpts_obj ·w kobj +L cls ·w cls +L dfl ·w dfl ,

[0111] where w box is the corresponding weight of L bbox , w pose is the corresponding weight of L kpts , w kobj is the corresponding weight of L kpts_obj , w cls is the corresponding weight of L cfs , and w dfl is the corresponding weight of L dfl .

[0112] The embodiment uses remote sensing images respectively containing corn plants or interference plants of different sizes for testing, and the results show that the improved YOLOv10 model can accurately predict the rotation bounding box and key points of the corn plants in different situations, as shown in Figures 5-7 .

[0113] Step S300, determining the corn emergence rate and emergence uniformity corresponding to the target area according to the rotation bounding box and key points corresponding to each of the corn plant images.

[0114] In short, each corn plant corresponds to a rotation bounding box, and then the number of corn emergence can be obtained according to the number of rotation bounding boxes, so as to calculate the corn emergence rate. The key points represent the center or center of the corn plant, and then the corn emergence uniformity can be obtained according to the uniformity of the key points.

[0115] In one implementation, determining the corn emergence rate corresponding to the target area according to the rotation bounding box corresponding to each of the corn plant images comprises:

[0116] Step S301, determining the number of rotation boxes corresponding to the target area according to the rotation bounding box corresponding to each of the corn plant images.

[0117] Step S302, obtaining the area of the target area, and determining the corn emergence rate of the target area according to the number of rotation boxes and the area of the target area.

[0118] Specifically, the embodiment calculates the number n of the rotation boxes corresponding to the target area according to the rotation bounding boxes corresponding to each corn plant image, and uses the number of rotation boxes to represent the corn emergence number. An actual area m corresponding to the target area or the remote sensing image is obtained, and the corn emergence rate corresponding to the target area or the remote sensing image is calculated according to the number of rotation boxes and the area:

[0119]

[0120] In an implementation manner, the emergence uniformity corresponding to the target area is determined according to the key points corresponding to each corn plant image, and the emergence uniformity corresponding to the target area is determined according to the key points corresponding to each corn plant image, including:

[0121] In step S303, the pixel distance between each adjacent corn plant is determined according to the key points corresponding to each corn plant image.

[0122] In step S304, a ground sampling distance is obtained, and the actual distance between each adjacent corn plant is determined according to each pixel distance and the ground sampling distance.

[0123] In step S305, the emergence uniformity corresponding to the target area is determined according to the actual distance between each adjacent corn plant.

[0124] Specifically, the embodiment counts the pixel distance between adjacent corn plants according to the key points (the center of each corn) corresponding to each corn plant image, and calculates the actual ground area corresponding to each pixel in the remote sensing image through the ground sampling distance (GSD). The ground sampling distance represents the actual distance corresponding to one pixel in the remote sensing image on the ground. If the GSD is 1 cm, it means that each pixel in the image represents an area of 1 square centimeter on the ground.

[0125]

[0126] where H is the flight height (the distance from the ground), S is the pixel size of the sensor, and f is the focal length of the sensor. If the pixel distance between the centers (key points) of two corns is d, the actual distance D between the two corns is:

[0127] D = d x GSD,

[0128] The emergence uniformity corresponding to the target area is determined according to the actual distance between each adjacent corn plant, including:

[0129] The average value μ of the actual distance between each adjacent corn plant is calculated, assuming that there are n pairs of actual distances between adjacent plants, which are respectively denoted as D1, D2, …, Dn. n The average value represents the average level of the actual distance between adjacent plants, and the calculation formula is:

[0130]

[0131] wherein μ is the mean value. D i is the actual distance between the i-th pair of adjacent corn plants. n is the number of pairs of adjacent corn plants.

[0132] The standard deviation σ, which measures the degree of dispersion of these distances, is calculated as follows:

[0133]

[0134] wherein σ is the standard deviation of the actual distances between the adjacent corn plants. D i is the actual distance between the i-th pair of adjacent corn plants. μ is the mean value. n is the number of pairs of adjacent corn plants.

[0135] The uniformity of emergence of the target area is calculated as follows:

[0136]

[0137] wherein CV is the coefficient of variation, σ is the standard deviation, and μ is the mean value.

[0138] In one implementation, the embodiments can be used to evaluate the emergence rate and uniformity of emergence of other crops, such as wheat, soybeans, etc., in addition to corn plants.

[0139] Based on the above embodiments, the present application also provides a device for evaluating the emergence rate and uniformity of emergence of corn, as shown in Figure 8 The device comprises:

[0140] The acquisition module 01 is configured to acquire a remote sensing image of a target area and determine a plurality of corn plant images based on the remote sensing image.

[0141] The prediction module 02 is configured to use an improved YOLOv10 model to predict the plurality of corn plant images and determine a rotation bounding box and a key point corresponding to each of the corn plant images, wherein the improved YOLOv10 model is realized by introducing a Mamba attention module into a neck network layer of the YOLOv10 model, and the key point is used to label the center of the corn plant.

[0142] The evaluation module 03 is configured to determine the emergence rate and uniformity of emergence of corn corresponding to the target area based on the rotation bounding box and the key point corresponding to each of the corn plant images.

[0143] Based on the above embodiments, the present application also provides a terminal, and a principle block diagram thereof can be as shown in Figure 9The terminal includes a processor, a memory, a network interface, and a display screen connected through a system bus. The processor of the terminal is configured to provide computing and control capabilities. The memory of the terminal includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement the method for evaluating corn emergence rate and emergence uniformity. The display screen of the terminal can be a liquid crystal display screen or an electronic ink display screen.

[0144] Those skilled in the art can understand that Figure 9 The principle block diagram shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the terminal to which the scheme of the present application is applied. The specific terminal can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0145] In an implementation manner, the memory of the terminal stores more than one program, and is configured to execute the more than one program by more than one processor, which includes instructions for performing the method for evaluating corn emergence rate and emergence uniformity.

[0146] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0147] In summary, the application discloses a corn emergence rate and emergence uniformity evaluation method, device, terminal and medium, the method determines a plurality of corn plant images according to a remote sensing image of a target area; an improved YOLOv10 model is used to predict the corresponding rotating bounding box and key point of each corn plant image, the key point is the center of the corn plant; the corn emergence rate and the emergence uniformity corresponding to the target area are determined according to the corresponding rotating bounding box and key point of each corn plant image. Since the improved YOLOv10 model is used to predict the rotating bounding box and key point of the corn plant in the application, the rotating bounding box can effectively reduce the interference of the background or other non-crop areas, improve the detection accuracy, and the key point can accurately mark the center of the corn plant, so the distance and distribution between plants can be more accurately evaluated, thereby effectively solving the problem that the center of the corn plant is marked by the center point of the detected target frame in the prior art, resulting in a large detection error.

[0148] It should be understood that the application is not limited to the above examples, and can be improved or changed according to the above description for those of ordinary skill in the art, and all these improvements and changes shall belong to the protection scope of the appended claims of the application.

Claims

1. A method for evaluating corn emergence rate and emergence uniformity, characterized by, The method comprises: acquiring a remote sensing image of a target area, and determining a plurality of corn plant images according to the remote sensing image; using an improved YOLOv10 model to perform prediction according to the plurality of corn plant images, and determining a rotation bounding box and a key point corresponding to each of the corn plant images, wherein the improved YOLOv10 model is realized by introducing a Mamba attention module into a neck network layer of a YOLOv10 model, and the key point is used to label the center of a corn plant; determining a corn emergence rate corresponding to the target area according to the rotation bounding box corresponding to each of the corn plant images, and determining an emergence uniformity corresponding to the target area according to the key point corresponding to each of the corn plant images; The Mamba attention module generates a feature representation x by adaptive pooling along a width dimension and a height dimension of an input tensor x and a feature representation x y , the input tensor being an input tensor corresponding to the Mamba attention module; respectively to the feature representation x x and the feature representation x y The width dimension and the height dimension respectively correspond to the dynamic feature weight determined by processing and nonlinear activation through the Mamba module. combining the input tensor with dynamic feature weights corresponding to the width dimension and the height dimension respectively through element-by-element multiplication to determine an output tensor of the Mamba attention module; determining the corn emergence rate corresponding to the target area according to the rotation bounding box corresponding to each of the corn plant images comprises: determining the number of rotation boxes corresponding to the target area according to the rotation bounding box corresponding to each of the corn plant images; acquiring an area of the target area, and determining the corn emergence rate of the target area according to the number of rotation boxes and the area; determining the emergence uniformity corresponding to the target area according to the key point corresponding to each of the corn plant images comprises: determining a pixel distance between adjacent corn plants according to the key point corresponding to each of the corn plant images; acquiring a ground sampling distance, and determining an actual distance between adjacent corn plants according to each of the pixel distances and the ground sampling distance; determining the emergence uniformity corresponding to the target area according to the actual distance between adjacent corn plants.

2. The method for evaluating corn emergence rate and emergence uniformity according to claim 1, characterized in that, The determination of the plurality of corn plant images according to the remote sensing image comprises: extracting inter-row straight lines of corn plants in the remote sensing image, and determining corn plant rows in the remote sensing image according to the inter-row straight lines; performing image segmentation on the remote sensing image according to the corn plant rows to determine a plurality of corn plant images.

3. The method for evaluating corn emergence rate and emergence uniformity according to claim 2, characterized in that, The extraction of the inter-row straight lines of corn plants in the remote sensing image comprises: performing image enhancement on the corn plants in the remote sensing image, and converting the enhanced remote sensing image into a binary image; performing morphological operation and Hough transformation on the binary image to extract the inter-row straight lines of corn plants in the remote sensing image.

4. The method of evaluating corn emergence rate and emergence uniformity according to claim 1, wherein, The training method of the improved YOLOv10 model comprises: acquiring a plurality of historical remote sensing images, performing rotation bounding box and key point labeling on each of the historical remote sensing images, and taking the labeled plurality of historical remote sensing images as training data; constructing a loss function corresponding to the improved YOLOv10 model; training the improved YOLOv10 model according to the training data and the loss function.

5. The method for evaluating corn emergence rate and emergence uniformity according to claim 4, characterized in that, The construction of the loss function corresponding to the improved YOLOv10 model comprises: calculating a classification loss, a bounding box loss and a key point loss in the training process of the improved YOLOv10 model; The classification loss, the bounding box loss, and the key point loss are weighted to determine a loss function corresponding to the improved YOLOv10 model.

6. A corn emergence rate and uniformity evaluation device characterized by, The device comprises: The acquisition module is configured to acquire a remote sensing image of a target region and determine a plurality of corn plant images based on the remote sensing image. The prediction module is configured to predict the plurality of corn plant images based on an improved YOLOv10 model to determine a rotation bounding box and key points corresponding to each of the corn plant images, wherein the improved YOLOv10 model is realized by introducing a Mamba attention module into a neck network layer of a YOLOv10 model, and the key points are used to label the center of a corn plant. The evaluation module is configured to determine a corn emergence rate of the target region based on the rotation bounding box corresponding to each of the corn plant images and determine an emergence uniformity of the target region based on the key points corresponding to each of the corn plant images. The Mamba attention module generates a feature representation x by adaptive pooling of an input tensor along a width dimension and a height dimension x and a feature representation x y , the input tensor being the input tensor corresponding to the Mamba attention module; respectively to the feature representation x x and the feature representation x y determined by the Mamba module through processing and nonlinear activation, the dynamic feature weight corresponding to the width dimension and the height dimension, respectively; The output tensor of the Mamba attention module is determined by combining the input tensor with dynamic feature weights corresponding to the width dimension and the height dimension, respectively, through element-wise multiplication. The corn emergence rate of the target region is determined based on the rotation bounding box corresponding to each of the corn plant images, including: The number of rotation bounding boxes corresponding to the target region is determined based on the rotation bounding box corresponding to each of the corn plant images. The area of the target region is acquired, and the corn emergence rate of the target region is determined based on the number of rotation bounding boxes and the area of the target region. The emergence uniformity of the target region is determined based on the key points corresponding to each of the corn plant images, including: The pixel distance between adjacent corn plants is determined based on the key points corresponding to each of the corn plant images. The ground sampling distance is acquired, and the actual distance between adjacent corn plants is determined based on the pixel distance and the ground sampling distance. The emergence uniformity of the target region is determined based on the actual distance between adjacent corn plants.

7. A terminal, characterized by comprising: The terminal comprises a memory and one or more processors; the memory stores one or more programs; the programs contain instructions for executing the corn emergence rate and emergence uniformity evaluation method according to any one of claims 1-5; and the processor is configured to execute the programs.

8. A computer-readable storage medium having a plurality of instructions stored thereon, characterized in that: The instructions are suitable for being loaded and executed by the processor to implement the steps of the corn emergence rate and emergence uniformity evaluation method according to any one of claims 1-5.

Images