Tobacco plant counting and plot identification method, device and equipment and storage medium

Abstract

The invention relates to the technical field of tobacco planting recognition, and discloses a tobacco plant counting and plot recognition method, device and equipment and a storage medium, and the method comprises the steps: obtaining a sample plot collection image of a tobacco planting sample plot region, carrying out the tobacco plant marking and tobacco plot marking, and constructing a training data set; training the pre-trained Swin Transformer model to obtain a tobacco plot recognition model and a tobacco point plant recognition model, and inputting the recognition acquisition image of the tobacco planting recognition area into the tobacco plot recognition model to obtain a tobacco plot recognition result, and in consideration of acquisition heights of different identification acquisition images, image processing and identification efficiency corresponding to the number of the identification acquisition images and area complete coverage conditions, determining an optimal identification acquisition image sample and inputting the optimal identification acquisition image sample into the tobacco point plant identification model to obtain a final tobacco planting identification result of the grass planting identification area. On the basis of ensuring the identification accuracy and the identification efficiency, the identification of the area of each tobacco plot and the tobacco point plant is realized.

Description

Technical Field

[0001] The invention relates to the technical field of tobacco planting identification, and in particular to a tobacco plant counting and plot identification method, device, equipment and storage medium. Background Art

[0002] In recent years, the application of deep learning and computer vision technology in the agricultural field has made significant progress. Especially in the tobacco planting industry, the efficiency and accuracy of production management can be greatly improved through automated and intelligent image analysis technology. Taking the intelligent management of tobacco planting areas in Yunnan Province as an example, the main objects of protection in tobacco planting areas in Yunnan Province are high-quality tobacco varieties and their growth environment. In addition, there is also the management of pests and diseases and monitoring of other agricultural ecological factors. The number of tobacco plants grown in the planting area each year and the health of tobacco are the most concerned issues for managers in the planting area and my country's agricultural departments. In order to better protect tobacco crops, improve planting efficiency, and reduce the impact of pests and diseases and adverse environmental factors on tobacco growth, it is essential for the planting area to apply technologies such as the Internet of Things and intelligent recognition to tobacco point plant identification and plot management.

[0003] In tobacco growing areas, by deploying IoT devices such as soil moisture sensors, temperature sensors, and light sensors, soil and climate conditions can be monitored in real time to provide optimal environmental data for tobacco growth. At the same time, using intelligent recognition technologies such as image recognition and machine learning algorithms, tobacco plant sites can be accurately identified, including plant count, growth status assessment, and pest and disease detection. Through intelligent identification of plots, managers can quickly locate problem areas and take corresponding agricultural measures in a timely manner, such as irrigation, fertilization, or spraying pesticides, to ensure the healthy growth of tobacco crops. In addition, through plot identification technology, planting areas can be accurately divided and managed to achieve the goal of precision agriculture. The specific conditions of each plot and the growth status of tobacco can be recorded and analyzed separately, thereby providing customized management strategies for the planting area. This intelligent management method not only improves the yield and quality of tobacco crops, but also helps save resources, reduce the impact on the environment, and promote the development of sustainable agriculture. In short, the application of IoT and intelligent recognition technology to tobacco plant site identification and plot management is of great significance for improving the efficiency and quality of tobacco planting, protecting the agricultural ecological environment, and realizing precision agriculture.

[0004] With the continuous development of precision agriculture, higher requirements are put forward for the point counting and plot identification of tobacco crops. Although the existing technologies have achieved remarkable results in their respective fields, there is still room for further exploration in the field of tobacco planting, especially in the refined management of point-level plants. For example, in 202410141779.2, a method for calculating the age of fruit trees in cloudy and rainy areas based on multi-source remote sensing images, multi-source high-resolution remote sensing images are used to extract the plot boundaries and identify orchard plots in the target area, and the breakpoint detection of the vegetation index of the orchard plots over the years is performed to calculate the age of the fruit trees, thereby solving the problem that the existing tree age measurement methods based on remote sensing technology are difficult to apply in cloudy and rainy areas. However, in nature, there are many types of crops with great morphological differences. In the study of crop recognition algorithms, the same set of algorithms applied to the recognition of large species may lead to low recognition accuracy. When the data set changes or there are few training base models for the data set in the process of specific species recognition, inaccurate recognition will occur.

[0005] At the same time, since the tobacco point strain technology and plot identification are mainly based on the tobacco planting area images collected by drones, in order to obtain complete data, redundant image collection is usually performed (for example, multiple image coverage collections at different positions and heights are performed for the tobacco planting area). Eventually, a large number of tobacco planting area images will be obtained, and each tobacco planting area image contains several complete or incomplete tobacco plots. Images at different positions and heights have different effects on tobacco planting identification (images at different positions will result in different splicing strategies when finally used to splice the entire tobacco planting area, which will affect the recognition efficiency; and images at different heights, on the one hand, affect the recognition accuracy due to the different quality of image pixel features, and on the other hand, different splicing strategies are used due to different covered areas, which affects the recognition efficiency). How to realize the recognition of the area and tobacco point strains of each tobacco plot on the basis of ensuring the recognition accuracy and recognition efficiency from a large number of tobacco planting area images is a key difficulty in this field.

[0006] Therefore, how to improve the efficiency and accuracy of tobacco plant counting and plot identification is a technical problem that needs to be solved urgently. Summary of the invention

[0007] The main purpose of the present invention is to provide a tobacco plant counting and plot identification method, device, equipment and storage medium, aiming to solve the current technical problems of low efficiency and accuracy of tobacco plant counting and plot identification.

[0008] To achieve the above object, the present invention provides a method for counting tobacco plants and identifying plots, comprising the following steps:

[0009] Acquire sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocess and data annotate the sample plot collection images, and obtain a training data set; wherein the data annotation includes annotating tobacco points and tobacco plots on the preprocessed sample plot collection images;

[0010] Inputting the training data set into a pre-trained Swin Transformer model to perform tobacco planting recognition training to obtain a tobacco planting recognition model;

[0011] Acquire a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image, input each identification acquisition image into the tobacco planting identification model, and obtain tobacco plot identification results and tobacco point plant identification results of each identification acquisition image;

[0012] Based on the tobacco plot recognition results and tobacco plant recognition results of each recognition acquisition image, and using the acquisition position information of each recognition acquisition image, the tobacco planting recognition results of the tobacco planting recognition area are generated.

[0013] Optionally, obtaining sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocessing and data labeling the sample plot collection images, and obtaining a training data set step specifically includes:

[0014] Driving a drone equipped with an image acquisition device to regularly perform a cruise flight over a tobacco planting sample plot area according to a preset route and controlling the image acquisition device to acquire images of the sample plot, thereby obtaining sample plot acquisition images of the tobacco planting sample plot area at different tobacco production stages;

[0015] The acquired sample plot images were preprocessed including image correction, image denoising and image enhancement. Rectangular frame annotation tool and polygon semi-automatic annotation tool were used to perform tobacco point and tobacco plot annotation on the preprocessed sample plot images respectively to obtain the training data set.

[0016] Optionally, the training data set is input into a pre-trained Swin Transformer model for tobacco planting identification training to obtain a tobacco planting identification model step, specifically including:

[0017] Constructing a Swin Transformer model framework, and pre-training the Swin Transformer model framework using a public data set to obtain a pre-trained Swin Transformer model;

[0018] The sample plot images for tobacco plot labeling and tobacco point strain labeling in the training data set are respectively input into the pre-trained Swin Transformer model to obtain a trained tobacco plot recognition model and a tobacco point strain recognition model.

[0019] Optionally, the steps of obtaining a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image, inputting each identification acquisition image into the tobacco planting identification model, and obtaining the tobacco plot identification result and tobacco point and strain identification result of each identification acquisition image specifically include:

[0020] Driving a drone equipped with an image acquisition device and a positioning device to perform a cruise flight over a tobacco planting identification area, controlling the image acquisition device and the positioning device to respectively perform multiple area coverage acquisitions of sample plot images and acquisition of acquisition position information, and obtaining a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image;

[0021] Each recognition acquisition image is input into the tobacco plot recognition model and the tobacco point strain recognition model for recognition, and based on the collection position information of each recognition acquisition image, the tobacco plot recognition result and the tobacco point strain recognition result of each recognition acquisition image are obtained.

[0022] Optionally, the multiple regional coverage acquisitions include several regional coverage acquisitions at different acquisition heights; inputting each recognition acquisition image into a tobacco plot recognition model and a tobacco point strain recognition model for recognition, and obtaining several tobacco plots and point strain data of each tobacco plot in each recognition acquisition image based on the acquisition position information of each recognition acquisition image, specifically including:

[0023] The acquisition height of each area coverage acquisition is added to each identification acquisition image acquired, and each identification acquisition image with the acquisition height is input into the tobacco plot recognition model to obtain the tobacco plot recognition result of each identification acquisition image; wherein the tobacco plot recognition result includes the tobacco plot boundary composed of a plurality of boundary contour points;

[0024] Calculate the region coordinate range of each tobacco plot in each recognition acquisition image according to the acquisition position information of each recognition acquisition image, the scale information of the image acquisition device when taking the image, and the image coordinates of each boundary contour point in the boundary of the tobacco plot;

[0025] Determine the best recognition acquisition image sample for performing tobacco point plant recognition based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image, and the regional coordinate range of the tobacco planting recognition area;

[0026] Several recognition collection images in the best recognition collection image samples are input into the tobacco plant strain recognition model for recognition, and the tobacco plant strain recognition result of each recognition collection image is obtained; wherein the tobacco plant strain recognition result includes the coordinate position of the tobacco plant strain in each tobacco plot.

[0027] Optionally, based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image, and the regional coordinate range of the tobacco planting recognition area, the step of determining the best recognition acquisition image sample for performing tobacco point plant recognition specifically includes:

[0028] Based on the regional coordinate range of each tobacco plot in each identification acquisition image, the acquisition height of each identification acquisition image and the regional range coordinates of the tobacco planting identification area;

[0029] The first constraint condition is that the union of the regional coordinate ranges of the tobacco plots of all the recognized collected images in the selected best recognized collected image samples covers the regional coordinate range of the tobacco planting recognition area; the second constraint condition is that the weight of each recognized collected image in the selected best recognized collected image samples is the ratio of its tobacco plot area to the tobacco plot area in all recognized collected images and the sum is lower than a preset value; the optimization goal is to minimize the number of recognized collected images in the selected best recognized collected image samples, and determine the best recognized collected image samples for performing tobacco point plant recognition.

[0030] Optionally, based on the tobacco plot recognition result and the tobacco plant recognition result of each recognition acquisition image, using the acquisition position information of each recognition acquisition image, the step of generating the tobacco planting recognition result of the tobacco planting recognition area specifically includes:

[0031] Based on the regional coordinate range of each tobacco plot in each recognition acquisition image in the best recognition acquisition image sample, performing the splicing of the regional coordinate range and deleting the duplicated areas, and determining the regional coordinate range and regional area of ​​each tobacco plot in the tobacco planting recognition area;

[0032] The number of tobacco plants in each tobacco plot is determined based on the coordinate positions of the tobacco plants in each tobacco plot in each recognition collection image in the best recognition collection image sample.

[0033] In addition, in order to achieve the above purpose, a tobacco plant counting and plot identification device includes:

[0034] An acquisition module is used to acquire sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocess and data annotate the sample plot collection images, and obtain a training data set; wherein the data annotation includes annotating tobacco points and tobacco plots on the preprocessed sample plot collection images;

[0035] A training module, used for inputting the training data set into a pre-trained Swin Transformer model to perform tobacco planting recognition training to obtain a tobacco planting recognition model;

[0036] The recognition module is used to obtain a plurality of recognition acquisition images of the tobacco planting recognition area and the acquisition position information of each recognition acquisition image, input each recognition acquisition image into the tobacco planting recognition model, and obtain the tobacco plot recognition result and tobacco point plant recognition result of each recognition acquisition image;

[0037] The generation module is used to generate tobacco planting identification results of the tobacco planting identification area based on the tobacco plot identification results and tobacco point plant identification results of each identification collection image and using the collection position information of each identification collection image.

[0038] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a tobacco strain counting and plot identification device, which includes: a memory, a processor, and a tobacco strain counting and plot identification program stored in the memory and executable on the processor, and when the tobacco strain counting and plot identification program is executed by the processor, the steps of the tobacco strain counting and plot identification method described above are implemented.

[0039] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a storage medium, on which a tobacco plant counting and plot identification program is stored. When the tobacco plant counting and plot identification program is executed by a processor, the steps of the above-mentioned tobacco plant counting and plot identification method are implemented.

[0040] The beneficial effects of the present invention are as follows: a method, device, equipment and storage medium for counting tobacco points and identifying plots are proposed, sample plot acquisition images of a tobacco planting sample plot area are acquired, a training data set is constructed after tobacco points and tobacco plots are labeled respectively, a pre-trained Swin Transformer model is trained using the training data set to obtain a tobacco plot recognition model and a tobacco point and strain recognition model, recognition acquisition images of a tobacco planting recognition area that are covered with different acquisition heights are acquired several times, the recognition acquisition images are input into the tobacco plot recognition model, and a tobacco plot recognition result of each recognition acquisition image is obtained, and according to the tobacco plot recognition result, the best recognition acquisition image sample is determined considering the acquisition heights of different recognition acquisition images, the image processing and recognition efficiency corresponding to the number of recognition acquisition images, and the condition of complete regional coverage, each recognition acquisition image in the best recognition acquisition image sample is input into the tobacco point and strain recognition model, and the final tobacco planting recognition result of the grass planting recognition area is obtained, and the area and tobacco points of each tobacco plot are recognized on the basis of ensuring the recognition accuracy and recognition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 The figure is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0043] Figure 2 The figure is a flow chart of an embodiment of the tobacco plant counting and plot identification method of the present invention.

[0044] Figure 3 Schematic diagram of the shift window method of the Swin Transformer architecture adopted in the present invention.

[0045] Figure 4 Schematic diagram of the structure of the Transformer module in the Swin Transformer of the present invention.

[0046] Figure 5 Schematic diagram of the hierarchical structure in the Swin Transformer of the present invention.

[0047] Figure 6 Schematic diagram of tobacco plot recognition results in an embodiment of the present invention.

[0048] Figure 7 Schematic diagram of tobacco strain identification results in an embodiment of the present invention.

[0049] Figure 8 The present invention is a structural block diagram of a tobacco plant counting and plot identification device in an embodiment of the present invention.

[0050] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, Figure 1 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0053] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0054] Those skilled in the art will understand that Figure 1 The structure of the device shown in the figure does not constitute a limitation of the device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0055] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a tobacco plant counting and plot identification program.

[0056] exist Figure 1 In the terminal shown, the network interface 1004 is mainly used to connect to the backend server and perform data communication with the backend server; the user interface 1003 is mainly used to connect to the client (user end) and perform data communication with the client; and the processor 1001 can be used to call the tobacco plant counting and plot identification program stored in the memory 1005, and perform the following operations:

[0057] Acquire sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocess and data annotate the sample plot collection images, and obtain a training data set; wherein the data annotation includes annotating tobacco points and tobacco plots on the preprocessed sample plot collection images;

[0058] Inputting the training data set into a pre-trained Swin Transformer model to perform tobacco planting recognition training to obtain a tobacco planting recognition model;

[0059] Acquire a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image, input each identification acquisition image into the tobacco planting identification model, and obtain tobacco plot identification results and tobacco point plant identification results of each identification acquisition image;

[0060] Based on the tobacco plot recognition results and tobacco plant recognition results of each recognition acquisition image, and using the acquisition position information of each recognition acquisition image, the tobacco planting recognition results of the tobacco planting recognition area are generated.

[0061] The specific embodiments of the present invention applied to the device are basically the same as the embodiments of the tobacco plant counting and plot identification method described below, and will not be described in detail here.

[0062] The embodiment of the present invention provides a method for counting tobacco plants and identifying plots. Figure 2 , Figure 2 The figure is a flow chart of an embodiment of the tobacco plant counting and plot identification method of the present invention.

[0063] In this embodiment, a method for counting tobacco plants and identifying plots includes the following steps:

[0064] S100: acquiring sample plot images of a tobacco planting sample plot area at different tobacco production stages, preprocessing and data labeling the sample plot images, and obtaining a training data set; wherein the data labeling includes labeling tobacco points and tobacco plots on the preprocessed sample plot images;

[0065] S200: inputting the training data set into a pre-trained Swin Transformer model to perform tobacco planting recognition training to obtain a tobacco planting recognition model;

[0066] S300: Acquire a plurality of recognition acquisition images of a tobacco planting recognition area and acquisition position information of each recognition acquisition image, input each recognition acquisition image into the tobacco planting recognition model, and obtain a tobacco plot recognition result and a tobacco point plant recognition result of each recognition acquisition image;

[0067] S400: Based on the tobacco plot recognition result and the tobacco plant recognition result of each recognition acquisition image, and using the acquisition position information of each recognition acquisition image, a tobacco planting recognition result of the tobacco planting recognition area is generated.

[0068] It should be noted that with the continuous development of precision agriculture, higher requirements are put forward for the point counting and plot identification of tobacco crops. Although the existing technologies have achieved remarkable results in their respective fields, there is still room for further exploration in the field of tobacco planting, especially in the refined management of point-level plants. For example, in 202410141779.2, a method for calculating the age of fruit trees in cloudy and rainy areas based on multi-source remote sensing images, multi-source high-resolution remote sensing images are used to extract the plot boundaries and identify orchard plots in the target area, and the breakpoint detection of the vegetation index of the orchard plots over the years is performed to calculate the age of the fruit trees, thereby solving the problem that the existing tree age measurement methods based on remote sensing technology are difficult to apply in cloudy and rainy areas. However, in nature, there are many types of crops with great morphological differences. In the study of crop recognition algorithms, the same set of algorithms applied to the recognition of large species may lead to low recognition accuracy. When the data set changes or there are few training base models for the data set in the process of specific species recognition, inaccurate recognition will occur. At the same time, since the tobacco point strain technology and plot identification are mainly based on the tobacco planting area images collected by drones, in order to obtain complete data, redundant image collection is usually performed (for example, multiple image coverage collections at different positions and heights are performed for the tobacco planting area). Eventually, a large number of tobacco planting area images will be obtained, and each tobacco planting area image contains several complete or incomplete tobacco plots. Images at different positions and heights have different effects on tobacco planting identification (images at different positions will result in different splicing strategies when finally used to splice the entire tobacco planting area, which will affect the recognition efficiency; and images at different heights, on the one hand, affect the recognition accuracy due to the different quality of image pixel features, and on the other hand, different splicing strategies are used due to different covered areas, which affects the recognition efficiency). How to realize the recognition of the area and tobacco point strains of each tobacco plot on the basis of ensuring the recognition accuracy and recognition efficiency from a large number of tobacco planting area images is a key difficulty in this field.

[0069] In order to solve the above problems, this embodiment obtains sample plot acquisition images of the tobacco planting sample plot area, constructs a training data set after respectively marking tobacco points and tobacco plots, uses the training data set to train the pre-trained SwinTransformer model to obtain a tobacco plot recognition model and a tobacco point strain recognition model, obtains recognition acquisition images that perform several regional coverage acquisitions at different acquisition heights on the tobacco planting recognition area, inputs the recognition acquisition images into the tobacco plot recognition model, obtains the tobacco plot recognition result of each recognition acquisition image, and determines the best recognition acquisition image sample according to the tobacco plot recognition result, considering the acquisition heights of different recognition acquisition images, the image processing and recognition efficiency corresponding to the number of recognition acquisition images, and the complete coverage condition of the area, and inputs each recognition acquisition image in the best recognition acquisition image sample into the tobacco point strain recognition model to obtain the final tobacco planting recognition result of the grass planting recognition area, so as to realize the recognition of the area and tobacco points of each tobacco plot on the basis of ensuring the recognition accuracy and recognition efficiency.

[0070] In a preferred embodiment, the steps of obtaining sample plot images of a tobacco planting sample plot area at different tobacco production stages, preprocessing and data annotation of the sample plot images, and obtaining a training data set specifically include:

[0071] S110: driving a drone equipped with an image acquisition device to regularly perform a cruise flight over a tobacco planting sample plot area according to a preset route and controlling the image acquisition device to acquire images of the sample plot, thereby obtaining sample plot acquisition images of the tobacco planting sample plot area at different tobacco production stages;

[0072] S120: performing preprocessing including image correction, image denoising and image enhancement on the acquired sample plot collection images, and using a rectangular frame annotation tool and a polygon semi-automatic annotation tool to respectively perform tobacco point and tobacco plot annotation on the sample plot collection images after the preprocessing, to obtain a training data set.

[0073] In this embodiment, in order to ensure the efficiency of drone data collection and produce high-standard drone real-life photo data to accurately reflect the actual situation. This embodiment plans to invest 27 drones, and actually invests 33 drones, including 20 sets of Mavic2 pro without screen version, 7 sets of Mavic2 zoom without screen version, 6 sets of phantom4 prov2.0 version; 20 phantom4 intelligent flight batteries, 2 sets of phantom4 intelligent chargers, 100 Mavic2 intelligent flight batteries, 22 sets of Mavic2 charging butlers, 11 sets of Mavic2 portable car chargers, 10 outdoor mobile power supplies, and 1 set of high-power generators, to obtain high-resolution tobacco planting area image data, which is transmitted to the ground control station via a wireless network, or downloaded through a storage device after the drone returns.

[0074] After that, the collected images are corrected, denoised and enhanced to ensure image quality for subsequent processing. Image correction: Use geometric transformation algorithms to correct distortion and tilt in the image to ensure the geometric accuracy of the image. Image denoising: Apply denoising algorithms such as median filtering and mean filtering to remove random noise and environmental interference in the image and enhance the clarity of the image. Image enhancement: Use image enhancement techniques such as histogram equalization to improve the contrast and brightness of the image and highlight the details of the tobacco plants.

[0075] Finally, the tobacco point strain counting and labeling software: labelimg.exe and the tobacco plot identification and labeling software: AnyLabeling-GPU.exe are used to realize the data labeling of the sample plot collection images. The data set of this embodiment contains a total of 203,467 images, which comprehensively covers all key stages of the tobacco growth process. These images carefully record the growth of tobacco in the seedling stage, root extension stage, vigorous growth stage and maturity stage. The amount of image data in each stage is relatively balanced, ensuring that the machine learning model can fully learn and understand the characteristic changes of tobacco at different growth stages. This diversity and richness of the data set is crucial for the training of the model, which helps to improve the accuracy and generalization ability of the model, so as to better apply it to the actual tobacco growth monitoring and analysis tasks. Tobacco point strain counting and labeling specifications: When performing tobacco point strain counting and labeling, the data standard specifications should be strictly followed, and the rectangular frame labeling tool should be used. During the labeling process, it must be ensured that each rectangular frame completely contains the entire tobacco point strain to avoid any part being omitted or truncated. In order to ensure the consistency and accuracy of the data, the operator needs to carefully check each labeling to prevent mislabeling. In addition, the labeling quality audit should be conducted regularly to ensure the high standard and high quality of the labeling data. Tobacco plot identification specifications: When identifying and marking tobacco plots, the polygon semi-automatic annotation tool should be used in accordance with data standard specifications. When marking, it is important to ensure that the boundaries of the entire plot are fully depicted and completely contained in the annotation area to prevent the edges of the plot from being missed or incompletely marked. The polygon annotation tool can provide a more detailed boundary description, but the operator must be careful to avoid mislabeling. To ensure the accuracy and consistency of the data, the annotation results should be regularly reviewed and verified to ensure that all plot annotations meet high standards of quality requirements.

[0076] In a preferred embodiment, the training data set is input into a pre-trained Swin Transformer model for tobacco planting identification training to obtain a tobacco planting identification model step, specifically comprising:

[0077] S210: constructing a Swin Transformer model framework, and pre-training the Swin Transformer model framework using a public data set to obtain a pre-trained Swin Transformer model;

[0078] S220: Inputting the sample plot images for tobacco plot labeling and tobacco point strain labeling in the training data set into the pre-trained Swin Transformer model respectively to obtain a trained tobacco plot recognition model and tobacco point strain recognition model.

[0079] In this embodiment, Swin Transformer is used to perform pre-training on public datasets such as COCO 2017 to obtain a powerful pre-trained model. Swin Transformer is an advanced architecture based on visual transformers, which has attracted much attention due to its significant improvements in computational efficiency and model performance. In the pre-training stage, the model can capture rich image information and semantic features by learning the feature representation of large-scale datasets. After that, the sample plot collection images for tobacco plot annotation and tobacco point strain annotation in the training dataset are respectively input into the pre-trained Swin Transformer model to obtain the trained tobacco plot recognition model and tobacco point strain recognition model.

[0080] It should be noted that Swin Transformer (Shifted Window Transformer) is a new type of visual Transformer architecture. By introducing hierarchical feature maps and window shifting strategies, it solves the computational complexity and scale change problems of traditional Transformer when applied in the visual field. The following is a detailed explanation of Swin Transformer:

[0081] 1. Hierarchical feature map: Swin Transformer starts with a small-sized image patch and merges adjacent image patches layer by layer to build a hierarchical feature map. Each image patch is regarded as a "token" whose feature is the connection of the original pixel RGB values. By merging layer by layer, the model can gradually capture a larger receptive field and adapt to visual elements of different scales.

[0082] 2. Window translation strategy: In traditional visual Transformers, the self-attention mechanism is calculated globally on the entire image, resulting in a quadratic increase in computational complexity as the image size increases. The window translation strategy proposed by Swin Transformer limits the self-attention calculation to non-overlapping local windows, thereby increasing the computational complexity linearly. Specifically, Swin Transformer translates the window divisions between consecutive self-attention layers, providing connections between the boundaries of the previous layer's windows, significantly enhancing the modeling capabilities of the model. Compared with the sliding window method, the window translation strategy significantly reduces computational latency while maintaining modeling capabilities.

[0083] 3. Specific implementation:

[0084] 3.1 Input processing: The input RGB image is split into non-overlapping image blocks. Assuming the image block size is 4×4, the feature dimension of each image block is 4×4×3=48. Then, these original features are projected to an arbitrary dimension (denoted as C) through a linear embedding layer for subsequent processing. Figure 3 As shown in the figure, the shifted window method used to calculate the self-attention mechanism in the Swin Transformer architecture is shown. Specifically, this method works through the following steps: Regular window division (layer l): In layer l, the image or feature map is divided into regular fixed-size windows. Self-attention is calculated independently within each window. The advantage of this is high computational efficiency because the calculations within each window are independent. Shifted window division (layer l+1): In layer l+1, the window division is shifted, which means that the position of the window is offset relative to the previous layer. The self-attention calculation of the new window crosses the window boundary of layer l. Self-attention calculation across window boundaries: This shifted window method enables the self-attention calculation in layer l+1 to cross the window boundary of layer l. This cross-window self-attention calculation provides connections between different windows, thereby enhancing the model's ability to capture global information.

[0085] 3.2 Transformer module: The Transformer module in Swin Transformer includes a multi-head self-attention mechanism and a feed-forward neural network (FFN). In each Transformer block, the self-attention mechanism is only calculated within a local window, rather than globally, thereby reducing computational complexity. Figure 4 and Figure 5As shown in Figure 1, in the Swin Transformer architecture, the model is usually composed of multiple Swin Transformer blocks, which are the basic building blocks of the model. Each Swin Transformer block includes two layers of consecutive sub-blocks, respectively called Swin Transformer layers.

[0086] Specifically, the workflow of two consecutive layers of Swin Transformer blocks is as follows: (1) The first layer of Swin Transformer blocks: Window partitioning: First, the input feature map is divided into windows of fixed size. Self-attention is calculated independently in each window. Self-attention calculation within the window: Self-attention is calculated within each window to capture local information within the window. (2) The second layer of Swin Transformer blocks: Window partition shifting: In this layer, the window partitioning is shifted relative to the first layer. This means that the new window position will cross the window boundary of the first layer. Self-attention calculation within the shifted window: Self-attention is calculated again in the shifted window, and this time the calculation crosses the boundary of the previous window, thereby capturing global information across windows.

[0087] 3.3 Hierarchical structure: Swin Transformer implements a hierarchical structure by merging image blocks layer by layer. The output feature map of each layer is reduced in size by downsampling (such as 2×2 pooling), so that the model can process larger input images while maintaining low computational complexity.

[0088] In a preferred embodiment, the steps of obtaining a plurality of identification acquisition images of a tobacco planting identification area and acquisition position information of each identification acquisition image, inputting each identification acquisition image into the tobacco planting identification model, and obtaining tobacco plot identification results and tobacco point strain identification results of each identification acquisition image specifically include:

[0089] S310: driving a drone equipped with an image acquisition device and a positioning device to perform a cruise flight over a tobacco planting identification area, controlling the image acquisition device and the positioning device to respectively perform multiple area coverage acquisitions of sample plot images and acquisition of acquisition position information, and obtaining a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image;

[0090] S320: Input each recognition acquisition image into the tobacco plot recognition model and the tobacco point strain recognition model for recognition, and obtain the tobacco plot recognition result and the tobacco point strain recognition result of each recognition acquisition image based on the collection position information of each recognition acquisition image.

[0091] Furthermore, the multiple regional coverage acquisitions include several regional coverage acquisitions at different acquisition heights; each recognition acquisition image is input into a tobacco plot recognition model and a tobacco point strain recognition model for recognition, and based on the acquisition position information of each recognition acquisition image, several tobacco plots in each recognition acquisition image and the point strain data of each tobacco plot are obtained, specifically including:

[0092] S321: attaching the acquisition height of each area coverage acquisition to each acquired recognition acquisition image, and inputting each recognition acquisition image with the acquisition height into the tobacco plot recognition model to obtain the tobacco plot recognition result of each recognition acquisition image; wherein the tobacco plot recognition result includes the tobacco plot boundary composed of a plurality of boundary contour points;

[0093] S322: Calculate and obtain the regional coordinate range of each tobacco plot in each recognition acquisition image according to the acquisition position information of each recognition acquisition image, the scale information of the image acquisition device when taking the image, and the image coordinates of each boundary contour point in the boundary of the tobacco plot;

[0094] S323: determining the best recognition acquisition image sample for performing tobacco point plant recognition based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image, and the regional coordinate range of the tobacco planting recognition area;

[0095] S324: Inputting several recognition collection images in the best recognition collection image samples into the tobacco strain recognition model for recognition, and obtaining tobacco strain recognition results for each recognition collection image; wherein the tobacco strain recognition results include the coordinate positions of the tobacco strains in each tobacco plot.

[0096] Furthermore, based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image, and the regional coordinate range of the tobacco planting recognition area, the steps of determining the best recognition acquisition image sample for performing tobacco point plant recognition specifically include:

[0097] S3231: based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image and the regional range coordinates of the tobacco planting recognition area;

[0098] S3232: The first constraint condition is that the union of the regional coordinate ranges of the tobacco plots of all the recognized collected images in the selected best recognized collected image samples covers the regional coordinate range of the tobacco planting recognition area; the second constraint condition is that the weight of each recognized collected image in the selected best recognized collected image samples is the ratio of its tobacco plot area to the tobacco plot area in all recognized collected images and the sum is lower than a preset value; the optimization goal is to minimize the number of recognized collected images in the selected best recognized collected image samples, and determine the best recognized collected image samples for performing tobacco plant identification.

[0099] Furthermore, based on the tobacco plot recognition result and the tobacco plant recognition result of each recognition acquisition image, the steps of generating the tobacco planting recognition result of the tobacco planting recognition area by using the acquisition position information of each recognition acquisition image specifically include:

[0100] S41: Based on the regional coordinate range of each tobacco plot in each recognition acquisition image in the best recognition acquisition image sample, performing splicing of the regional coordinate range and deletion of duplicate areas, and determining the regional coordinate range and regional area of ​​each tobacco plot in the tobacco planting recognition area;

[0101] S42: Determine the number of tobacco plants in each tobacco plot based on the coordinate positions of the tobacco plants in each tobacco plot in each recognition and acquisition image in the best recognition and acquisition image sample.

[0102] In this embodiment, by acquiring sample plot acquisition images of the tobacco planting sample plot area, a training data set is constructed after tobacco point strain labeling and tobacco plot labeling are respectively performed, and the pre-trained Swin Transformer model is trained using the training data set to obtain a tobacco plot recognition model and a tobacco point strain recognition model, and the recognition acquisition images of performing several times of area coverage acquisition at different acquisition heights on the tobacco planting recognition area are obtained, and the recognition acquisition images are input into the tobacco plot recognition model to obtain the tobacco plot recognition results of each recognition acquisition image (such as Figure 6 As shown in the figure), according to the tobacco plot recognition results, the best recognition acquisition image samples are determined by considering the acquisition heights of different recognition acquisition images, the image processing and recognition efficiency corresponding to the number of recognition acquisition images, and the complete coverage conditions of the area. Each recognition acquisition image in the best recognition acquisition image sample is input into the tobacco point plant recognition model to obtain the final tobacco planting recognition result of the grass planting recognition area (as shown in the figure). Figure 7 As shown in the figure), the area of ​​each tobacco plot and the tobacco plant can be identified on the basis of ensuring the recognition accuracy and recognition efficiency.

[0103] Reference Figure 8 , Figure 8 It is a structural block diagram of an embodiment of the tobacco plant counting and plot identification device of the present invention.

[0104] like Figure 8 As shown, the tobacco plant counting and plot identification device proposed in the embodiment of the present invention includes:

[0105] The acquisition module 10 is used to acquire sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocess and data annotate the sample plot collection images, and obtain a training data set; wherein the data annotation includes annotating tobacco points and tobacco plots on the preprocessed sample plot collection images;

[0106] A training module 20, for inputting the training data set into a pre-trained Swin Transformer model for tobacco planting identification training to obtain a tobacco planting identification model;

[0107] The recognition module 30 is used to obtain a plurality of recognition acquisition images of the tobacco planting recognition area and the acquisition position information of each recognition acquisition image, input each recognition acquisition image into the tobacco planting recognition model, and obtain the tobacco plot recognition result and tobacco point strain recognition result of each recognition acquisition image;

[0108] The generation module 40 is used to generate tobacco planting identification results of the tobacco planting identification area based on the tobacco plot identification results and tobacco point strain identification results of each identification collection image and using the collection position information of each identification collection image.

[0109] Other embodiments or specific implementations of the tobacco plant counting and plot identification device of the present invention can refer to the above-mentioned method embodiments, which will not be described in detail here.

[0110] In addition, the present invention also proposes a tobacco strain counting and plot identification device, which includes: a memory, a processor, and a tobacco strain counting and plot identification program stored in the memory and executable on the processor, and when the tobacco strain counting and plot identification program is executed by the processor, the steps of the tobacco strain counting and plot identification method described above are implemented.

[0111] The specific implementation of the tobacco plant counting and plot identification device of the present application is basically the same as the above-mentioned tobacco plant counting and plot identification method embodiments, and will not be repeated here.

[0112] In addition, the present invention also provides a readable storage medium, which includes a computer-readable storage medium on which a tobacco plant counting and plot identification program is stored. The readable storage medium may be Figure 1The memory 1005 in the terminal may also be at least one of a ROM (Read-Only Memory) / RAM (Random Access Memory), a magnetic disk, and an optical disk. The readable storage medium includes a number of instructions for enabling a tobacco plant counting and plot identification device with a processor to execute the tobacco plant counting and plot identification method described in each embodiment of the present invention.

[0113] The specific implementation of the readable storage medium of the present application is basically the same as the above-mentioned tobacco plant counting and plot identification method embodiments, and will not be repeated here.

[0114] It is understood that, in the description of this specification, the description with reference to the terms "one embodiment", "another embodiment", "other embodiments", or "first to Nth embodiments" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0115] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0116] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0117] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the present invention can still be modified or replaced by equivalents. Any modification or partial replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for counting tobacco plants and identifying plots, characterized in that: The following steps are involved: Acquire sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocess and data annotate the sample plot collection images, and obtain a training data set; wherein the data annotation includes annotating tobacco points and tobacco plots on the preprocessed sample plot collection images; Inputting the training data set into a pre-trained Swin Transformer model to perform tobacco planting recognition training to obtain a tobacco planting recognition model; Acquire a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image, input each identification acquisition image into the tobacco planting identification model, and obtain tobacco plot identification results and tobacco point plant identification results of each identification acquisition image; Based on the tobacco plot recognition results and tobacco plant recognition results of each recognition acquisition image, and using the acquisition position information of each recognition acquisition image, the tobacco planting recognition results of the tobacco planting recognition area are generated.

2. The tobacco plant counting and plot identification method according to claim 1, characterized in that: The steps of obtaining sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocessing and data labeling the sample plot collection images, and obtaining a training data set specifically include: Driving a drone equipped with an image acquisition device to regularly perform a cruise flight over a tobacco planting sample plot area according to a preset route and controlling the image acquisition device to acquire images of the sample plot, thereby obtaining sample plot acquisition images of the tobacco planting sample plot area at different tobacco production stages; The acquired sample plot images were preprocessed including image correction, image denoising and image enhancement. Rectangular frame annotation tool and polygon semi-automatic annotation tool were used to perform tobacco point and tobacco plot annotation on the preprocessed sample plot images respectively to obtain the training data set.

3. The tobacco plant counting and plot identification method according to claim 1, characterized in that: The training data set is input into the pre-trained Swin Transformer model to perform tobacco planting identification training to obtain a tobacco planting identification model step, specifically including: Constructing a Swin Transformer model framework, and pre-training the Swin Transformer model framework using a public data set to obtain a pre-trained Swin Transformer model; The sample plot images for tobacco plot labeling and tobacco point strain labeling in the training data set are respectively input into the pre-trained Swin Transformer model to obtain a trained tobacco plot recognition model and a tobacco point strain recognition model.

4. The tobacco plant counting and plot identification method according to claim 3, characterized in that: The steps of obtaining a plurality of identification acquisition images of a tobacco planting identification area and acquisition position information of each identification acquisition image, inputting each identification acquisition image into the tobacco planting identification model, and obtaining a tobacco plot identification result and a tobacco point plant identification result of each identification acquisition image specifically include: Driving a drone equipped with an image acquisition device and a positioning device to perform a cruise flight over a tobacco planting identification area, controlling the image acquisition device and the positioning device to respectively perform multiple area coverage acquisitions of sample plot images and acquisition of acquisition position information, and obtaining a plurality of identification acquisition images of the tobacco planting identification area and acquisition position information of each identification acquisition image; Each recognition acquisition image is input into the tobacco plot recognition model and the tobacco point strain recognition model for recognition, and based on the collection position information of each recognition acquisition image, the tobacco plot recognition result and the tobacco point strain recognition result of each recognition acquisition image are obtained.

5. The tobacco plant counting and plot identification method according to claim 4, characterized in that: The multiple regional coverage acquisitions include several regional coverage acquisitions at different acquisition heights; each recognition acquisition image is input into a tobacco plot recognition model and a tobacco point strain recognition model for recognition, and based on the acquisition position information of each recognition acquisition image, several tobacco plots in each recognition acquisition image and point strain data of each tobacco plot are obtained, specifically including: The acquisition height of each area coverage acquisition is added to each identification acquisition image acquired, and each identification acquisition image with the acquisition height is input into the tobacco plot recognition model to obtain the tobacco plot recognition result of each identification acquisition image; wherein the tobacco plot recognition result includes the tobacco plot boundary composed of a plurality of boundary contour points; Calculate the region coordinate range of each tobacco plot in each recognition acquisition image according to the acquisition position information of each recognition acquisition image, the scale information of the image acquisition device when taking the image, and the image coordinates of each boundary contour point in the boundary of the tobacco plot; Determine the best recognition acquisition image sample for performing tobacco point plant recognition based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image, and the regional coordinate range of the tobacco planting recognition area; Several recognition collection images in the best recognition collection image samples are input into the tobacco plant strain recognition model for recognition, and the tobacco plant strain recognition result of each recognition collection image is obtained; wherein the tobacco plant strain recognition result includes the coordinate position of the tobacco plant strain in each tobacco plot.

6. The tobacco plant counting and plot identification method according to claim 5, characterized in that: Based on the regional coordinate range of each tobacco plot in each recognition acquisition image, the acquisition height of each recognition acquisition image, and the regional coordinate range of the tobacco planting recognition area, the steps of determining the best recognition acquisition image sample for performing tobacco point plant recognition specifically include: Based on the regional coordinate range of each tobacco plot in each identification acquisition image, the acquisition height of each identification acquisition image and the regional range coordinates of the tobacco planting identification area; The first constraint condition is that the union of the regional coordinate ranges of the tobacco plots of all the recognized collected images in the selected best recognized collected image samples covers the regional coordinate range of the tobacco planting recognition area; the second constraint condition is that the weight of each recognized collected image in the selected best recognized collected image samples is the ratio of its tobacco plot area to the tobacco plot area in all recognized collected images and the sum is lower than a preset value; the optimization goal is to minimize the number of recognized collected images in the selected best recognized collected image samples, and determine the best recognized collected image samples for performing tobacco point plant recognition.

7. The tobacco plant counting and plot identification method according to claim 6, characterized in that: Based on the tobacco plot recognition result and tobacco plant recognition result of each recognition acquisition image, the steps of generating tobacco planting recognition result of the tobacco planting recognition area by using the acquisition position information of each recognition acquisition image specifically include: Based on the regional coordinate range of each tobacco plot in each recognition acquisition image in the best recognition acquisition image sample, performing the splicing of the regional coordinate range and deleting the duplicated areas, and determining the regional coordinate range and regional area of ​​each tobacco plot in the tobacco planting recognition area; The number of tobacco plants in each tobacco plot is determined based on the coordinate positions of the tobacco plants in each tobacco plot in each recognition collection image in the best recognition collection image sample.

8. A tobacco plant counting and plot identification device, characterized in that: include: An acquisition module is used to acquire sample plot collection images of tobacco planting sample plots at different tobacco production stages, preprocess and data annotate the sample plot collection images, and obtain a training data set; wherein the data annotation includes annotating tobacco points and tobacco plots on the preprocessed sample plot collection images; A training module, used for inputting the training data set into a pre-trained Swin Transformer model to perform tobacco planting recognition training to obtain a tobacco planting recognition model; The recognition module is used to obtain a plurality of recognition acquisition images of the tobacco planting recognition area and the acquisition position information of each recognition acquisition image, input each recognition acquisition image into the tobacco planting recognition model, and obtain the tobacco plot recognition result and tobacco point plant recognition result of each recognition acquisition image; The generation module is used to generate tobacco planting identification results of the tobacco planting identification area based on the tobacco plot identification results and tobacco point plant identification results of each identification collection image and using the collection position information of each identification collection image.

9. A tobacco plant counting and plot identification device, characterized in that: The tobacco strain counting and plot identification device comprises: a memory, a processor, and a tobacco strain counting and plot identification program stored in the memory and executable on the processor. When the tobacco strain counting and plot identification program is executed by the processor, the steps of the tobacco strain counting and plot identification method as described in any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: The storage medium stores a tobacco plant counting and plot identification program, which, when executed by a processor, implements the steps of the tobacco plant counting and plot identification method according to any one of claims 1 to 7.

Images