A method, system, device and storage medium for diagnosing and detecting missing plants

Abstract

This invention provides a method, system, device, and storage medium for diagnosing diseased plants and detecting missing plants, relating to the field of diseased plant diagnosis and missing plant detection. The method includes: acquiring target images of plants within a target area and generating target orthophotos based on the target images; dividing the target orthophotos into multiple target window images using a set sliding window; inputting the multiple target window images into a detection model to obtain detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease severity level of the plant; when the disease severity level is zero, it indicates a missing plant; the detection model is established using machine learning methods; this invention can quickly and accurately detect plants.

Description

Technical Field

[0001] This invention relates to the field of diseased plant diagnosis and missing plant detection, and in particular to a method, system, device and storage medium for diseased plant diagnosis and missing plant detection. Background Technology

[0002] China is a major grape producer in the world. With the increase in planting area and the age of the vines, the frequency and severity of diseases have gradually intensified, making diseases one of the most important factors restricting the sustainable development of the grape and wine industry.

[0003] Grapes are among the most virus-prone perennial fruit trees. To date, approximately 70 types of grape viruses and viroids have been reported worldwide. Grape leafroll disease (GLD) is the most serious viral disease affecting grapes, considered the most prevalent and economically damaging virus globally. GLD inhibits grape growth, shortens vine lifespan, reduces photosynthesis, and affects carbohydrate metabolism, particularly impacting red varieties. Economic losses due to yield reduction and fruit quality degradation caused by GLD infection are estimated at US$25,000 to US$40,000 per hectare.

[0004] As part of smart agriculture, refined vineyard management, particularly the management decisions aimed at reducing economic losses from grape leafroll disease, relies on continuous monitoring of the disease's severity. The spatial distribution of virus-carrying grape leafroll strains within the vineyard and the assessment of their severity are crucial for effective management decisions. Identification of virus-carrying strains primarily relies on polymerase chain reaction (PCR) detection, hyperspectral imaging, or manual field surveys. However, these methods, operating at the leaf scale, are ill-suited to the large-scale plantings of vineyards. Assessing the severity of leafroll disease across the entire vine still depends on manual surveys, relying on the number of leaves exhibiting symptoms. In practice, these methods only allow for sampling, preventing the acquisition of a comprehensive spatial distribution of virus-carrying strains across the entire vineyard, leading to inaccurate vineyard testing results.

[0005] Therefore, it is necessary to develop an automated field-scale solution for detecting grape leafroll disease, which can be practical, reliable, economical, efficient and accurate in detecting the spatial distribution of leafroll virus-carrying strains in vineyards, assessing the severity of the disease, and providing meaningful information for the decision-making process. Summary of the Invention

[0006] The purpose of this invention is to provide a method, system, device, and storage medium for diagnosing diseased plants and detecting missing plants, which can quickly and accurately detect plants.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A method for diagnosing diseased plants and detecting missing plants, the method comprising:

[0009] Acquire target images of plants within the target area, and generate target orthophotos based on the target images;

[0010] The target orthophoto image is divided into multiple target window images by using a set sliding window;

[0011] Multiple target window images are input into the detection model to obtain detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease severity of the plant;

[0012] When the disease severity level is zero, it indicates a lack of plants;

[0013] The detection model was established using machine learning methods.

[0014] Optionally, dividing the target orthophoto image into multiple target window images using a set sliding window specifically includes:

[0015] The set sliding window is moved on the target orthophoto according to the set motion trajectory to obtain multiple target window images; the set motion trajectory is determined by the starting position and the ending position of the target orthophoto.

[0016] Optionally, the size of the sliding window is determined based on the spacing between plants in the target area.

[0017] Optionally, the method for determining the detection model is as follows:

[0018] Acquire training data; the training data is obtained by augmenting the original training data using data augmentation methods; the original training data includes: original orthophotos of plants in different disease severity levels at each sampling point area within the experimental area.

[0019] Multiple training window images are obtained by moving a set sliding window over the orthophoto of the training data;

[0020] The multiple training window images are divided into a training set and a validation set;

[0021] Construct a shared neural network;

[0022] The training set is input into the shared neural network, and the parameters in the shared neural network are trained with the goal of minimizing the error, so as to obtain the trained shared neural network.

[0023] The parameters of the trained shared neural network are adjusted using the validation set to obtain the detection model.

[0024] Optionally, generating a target orthophoto image based on the target image specifically includes:

[0025] The target image is converted into a target orthophoto using a super-resolution network; the resolution of the target orthophoto is greater than a set resolution value; wherein the super-resolution network is determined using machine learning methods.

[0026] Optionally, the method for determining the super-resolution network is as follows:

[0027] Acquire training images; the training images include original images of plants in each sampling point area within the test area;

[0028] Construct a training network;

[0029] The training image is input into the training network, and the peak signal-to-noise ratio of the training network is adjusted with the goal of minimizing the loss, to obtain the adjusted training network.

[0030] The adjusted training network was determined to be a super-resolution network.

[0031] Optionally, the method further includes:

[0032] The detection information is classified and saved according to the severity of the disease.

[0033] A system for diagnosing diseased plants and detecting missing plants, the system comprising:

[0034] The image processing module is used to acquire target images of plants in the target area and generate target orthophotos based on the target images.

[0035] The target window image acquisition module is used to divide the target orthophoto into multiple target window images using a set sliding window;

[0036] The detection information determination module is used to input multiple target window images into the detection model to obtain detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease level of the plant;

[0037] When the disease severity level is zero, it indicates a lack of plants;

[0038] The detection model was established using machine learning methods.

[0039] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor running the computer program to enable the electronic device to perform any of the above-described methods for diagnosing diseased plants and detecting missing plants.

[0040] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the disease diagnosis and missing plant detection methods described in any one of the above-mentioned methods.

[0041] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0042] This invention provides a method, system, device, and storage medium for diagnosing diseased plants and detecting missing plants. The method acquires target images of plants within a target area and generates target orthophotos based on these images. A sliding window is used to divide the target orthophotos into multiple target window images. These multiple target window images are input into a detection model to obtain detection information corresponding to each target window image. The detection information includes the plant's geographical location and disease severity level. A disease severity level of zero indicates a missing plant. Because the sliding window and the detection model established based on machine learning methods are used to diagnose and detect plants within the target area, the diagnosis and detection are faster and more accurate. Therefore, this invention can quickly and accurately detect plants. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart of a method for diagnosing diseased plants and detecting missing plants provided in an embodiment of the present invention;

[0045] Figure 2 This is a structural diagram of the diseased plant diagnosis and missing plant detection system provided in an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of the parameters corresponding to each level of the detection model provided in the embodiments of the present invention.

[0047] Symbol explanation:

[0048] Image processing module-1; target window image acquisition module-2; detection information determination module-3. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] The purpose of this invention is to provide a method, system, device, and storage medium for diagnosing diseased plants and detecting missing plants. Remote sensing technology, with its rapid data acquisition, wide coverage, and timeliness, can quickly and extensively investigate leaf roll disease without being destructive. Plant disease and pest detection based on remote sensing technology can be considered a method of plant "radiodiagnosis," effectively providing non-contact and continuous spatial monitoring of diseases and pests. Early research and applications in this area can be traced back to the 1980s. Through visual interpretation of aerial or satellite imagery, areas damaged by plant diseases and pests can be identified. Remote sensing data can be acquired from multiple platforms, including near-ground remote sensing equipment, aerial, and satellite platforms. Depending on the altitude of different platforms and the type of sensor, multi-source remote sensing data with different spatial, temporal, spectral, and spatial resolutions at multiple scales can be obtained. With the rapid development of computer science and sensor technology, extensive remote sensing data has been used for field-scale plant disease and pest detection. In the past few decades, research on computer vision algorithms, feature extraction, and machine learning algorithms has been conducted at multiple scales, revealing new possibilities for monitoring plant diseases and pests across regions.

[0051] Understanding the connection between remote sensing data and plant pathology theory has broadened the scope of agricultural information acquisition in many ways, thus providing a foundation for effective decision-making in smart agriculture. Existing research not only focuses on detecting the occurrence of specific pests and diseases but also assesses their severity and maps their distribution at the plot or regional level. In some cases, remote sensing-based plant pest and disease detection results have begun to show great potential in practical applications, such as facilitating precise field spraying or supporting high-throughput phenotypic analysis of plants.

[0052] In recent years, the use of machine learning to process remote sensing data has provided new approaches to feature extraction and modeling for plant disease and pest detection, surpassing traditional spectral features and statistical discrimination methods. Computer vision technology combined with deep learning represents a cutting-edge technology for object identification and classification. Numerous studies have shown that combining computer vision technology and deep learning to process remote sensing imagery significantly improves the accuracy of plant disease and pest detection.

[0053] To achieve the detection and severity assessment of leafroll disease-carrying strains, UAV remote sensing, with its high spatial resolution and rich texture information, compensates for the low spatial resolution of satellite remote sensing data, making it more suitable for acquiring spatial distribution information of virus-carrying strains in large-scale vineyards. Combining computer vision technology and deep learning to process UAV remote sensing imagery can provide a solution for rapidly, efficiently, and accurately investigating the prevalence of leafroll disease and locating virus-carrying strains in wine grapes at the field scale. Therefore, this invention provides a method, system, device, and storage medium for diagnosing diseased plants and detecting missing plants, enabling rapid and accurate plant detection.

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1

[0056] like Figure 1 As shown in the figure, this invention provides a method for diagnosing diseased plants and detecting missing plants, the method comprising:

[0057] Step 100: Acquire target images of the plants within the target area and generate a target orthophoto based on the target images. A drone can be used to acquire the target images of the plants within the target area.

[0058] Specifically, generating a target orthophoto from a target image includes:

[0059] A super-resolution network is used to convert the target image into a target orthophoto; the resolution of the target orthophoto is greater than a set resolution value; the super-resolution network is determined using machine learning methods.

[0060] The method for determining this super-resolution network is as follows:

[0061] Acquire training images; training images include original images of plants in each sampling point area within the experimental area.

[0062] Construct a training network.

[0063] The training images are input into the training network, and the peak signal-to-noise ratio (PSNR) of the training network is adjusted with the goal of minimizing the loss, resulting in the adjusted training network. Minimizing the loss aims to minimize the loss function, which is the function corresponding to the mean squared error.

[0064] The adjusted training network was determined to be a super-resolution network.

[0065] Step 200: Divide the target orthophoto into multiple target window images using a set sliding window.

[0066] Specifically, step 200 includes:

[0067] The set sliding window moves on the target orthophoto according to the set motion trajectory to obtain multiple target window images; the set motion trajectory is determined by the start position and end position of the target orthophoto.

[0068] Furthermore, the size of the sliding window is determined based on the spacing between plants in the target area.

[0069] MATLAB programming can be used to process target orthophotos. The positions of the two endpoints of plants within the target area in the orthophoto can be obtained through human-computer interaction; these are the starting and ending positions of the orthophoto. The trajectory of the sliding window on the orthophoto can be a straight line determined by the positions of the two endpoints of the plants within the target area.

[0070] Furthermore, the sliding window size is set according to the spacing between plants within the target vineyard. Taking grapes grown in a vineyard as an example, the typical spacing between grapevines for wine grapes or regular grapes in a vineyard is 1 to 2 meters, so the pixel size of the sliding window is set to 96*96. In this example, the ground sampling distance of the orthophoto is 1.74cm, meaning one pixel corresponds to 1.74cm. Therefore, 96*96 pixels correspond to an actual distance of 167cm*167cm.

[0071] Step 300: Input multiple target window images into the detection model to obtain the detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease level of the plant.

[0072] When the disease severity level is zero, it indicates a lack of plants; the detection model is established using machine learning methods.

[0073] The method for determining the detection model is as follows:

[0074] Acquire training data; the training data is obtained by augmenting the original training data using data augmentation methods; the original training data includes: original orthophotos of plants in different disease severity levels in each sampling area within the experimental zone.

[0075] Multiple training window images are obtained by moving a set sliding window over the orthophoto of the training data.

[0076] Multiple training window images are divided into training and validation sets.

[0077] Construct a shared neural network.

[0078] The training set is input into the shared neural network, and the parameters in the shared neural network are trained with the goal of minimizing the error, resulting in the trained shared neural network.

[0079] The parameters of the trained shared neural network are adjusted using a validation set to obtain the detection model.

[0080] As an optional implementation, the method further includes:

[0081] The test information is categorized and saved according to the severity of the disease.

[0082] This involves using a detection model to examine images acquired through a sliding window, saving the center point of the sliding window based on the classification results, converting the image location into geographic coordinates using the geographic information of the orthophoto, and writing the data to a storage module (or an SHP file). This allows for visualization of the distribution of virus-carrying strains and the presence of missing strains in the vineyards. Taking a vineyard as an example, the virus-carrying strains could be grapevines in a vineyard with leaf curl disease. The detection model could be the SE-ResNet classification model.

[0083] The method for diagnosing diseased plants and detecting missing plants provided in this embodiment can also be implemented in the following steps in practical applications: Taking a vineyard as an example, the diseased plant is a leafroll virus-carrying strain.

[0084] Step 1: Collect drone data

[0085] A drone equipped with a visible light camera was used to acquire visible light images of the vineyard canopy, which are either photographs or target images. The drone operated in clear weather, with a flight path set to 85% longitudinal overlap and 75% lateral overlap, at an altitude of 60 meters. The acquired image data had a spatial resolution of 1.74 centimeters. The drone's battery life was sufficient to cover an area of ​​approximately 173 acres for this mission. Before the drone flight, a 0.2m x 0.2m white light reference board was used to obtain reference data for radiometric calibration to account for the effects of camera characteristics, reflectivity characteristics, and environmental variations.

[0086] Step 2: Processing Drone Images

[0087] The acquired UAV visible light image data, i.e., the target image, is processed through geometric correction, feature point matching, image stitching, and georegistration to generate an orthophoto of the wine grape canopy. The orthophoto contains information from four channels: R, G, B, and A. The A channel refers to transparency. In this example, the A channel is 100%, corresponding to a value of 255. Generating the target orthophoto from the target image can be achieved using Pix4D mapper software.

[0088] Step 3: Construct a diagnostic model for the wine grape canopy, i.e., construct a detection model.

[0089] This paper utilizes a deep learning classification algorithm to achieve accurate diagnosis of the severity of viral strains. The deep learning network architecture used is SE-ResNet18 with an added SE (Squeeze-and-Excitation Networks) module. The specific steps for building the diagnostic model are as follows:

[0090] Step 1: Building the dataset, i.e., obtaining training data

[0091] The dataset, or training data, includes orthophoto images and their corresponding classification labels. First, we need to obtain raw orthophotos of grapevines at different disease severity levels.

[0092] The disease status of grapevines at sampling points within the experimental area was assessed through field investigation, and the severity of disease in virus-carrying strains was recorded according to a 5-level standard. A handheld inspection device (with an accuracy of 0.01 meters) was used to record the geographical information of the sampling points, with the WGS-84 coordinate system used as the positioning coordinate system.

[0093] The criteria for assessing the severity of the disease are:

[0094] Grade 1: No symptoms; Grade 2: Mild leaf curling symptoms only on older leaves at the base of branches; Grade 3: Less than one-third of the plant's leaves show leaf curling symptoms; Grade 4: One-third to one-half of the plant's leaves show leaf curling symptoms; Grade 5: More than half of the plant's leaves show leaf curling symptoms.

[0095] The symptoms of leaf curl disease are that the leaf tissue of the diseased leaves is mottled with light red color, and then the color of the spots gradually darkens and the spots will spread to the entire leaf.

[0096] Using the geographic information of the sampling points, the corresponding image locations in the orthophotos were retrieved, and the center point of the corresponding virus-carrying strain area in the orthophotos was found. The sampling area size for each sampling point was 1.67 meters × 1.67 meters. The orthophotos were cropped according to the area location. The classification labels were derived from the severity of the virus-carrying strain recorded in the field survey corresponding to that sampling point.

[0097] The original orthophotos for each disease severity level consist of 600 images, with six classification labels: Level 0 (missing strain), Level 1 (carrying virus), Level 2 (carrying virus), Level 3 (carrying virus), Level 4 (carrying virus), and Level 5 (carrying virus). Sixty samples from each category are input into the original orthophotos for that disease severity level, and 29 samples are selected as the test set. Data augmentation methods are used to augment the original training data for each disease severity level; specifically, data augmentation methods including adding Gaussian noise are used to augment the data. Ultimately, each original orthophoto for each disease severity level contains 600 samples. The original orthophotos for all disease severity levels are combined to obtain the training data; then, 75% of this data is used as the training set, and 25% as the validation set.

[0098] Step 2: Image Super-Resolution Based on VDSR

[0099] The images in the dataset initially had a resolution of 96*96 pixels, which were super-resolved to 224*224 pixels using the VDSR method. VDSR is a super-resolution model that recovers high-frequency information from low-frequency information in images by training a 41-layer CNN network. Data sets with spatial resolutions of 0.69 cm and 5.34 cm were used as the high-resolution and low-resolution images for training the model. Georegistration was used to ensure consistent regions in the matched images, and bicubic interpolation was used to match the size of the low-resolution images to the high-resolution images. The input image pixel size was 96×96. The mean squared error (MSE) was used as the loss function to obtain the optimal peak signal-to-noise ratio (PSNR), and a total of 100 image pairs were trained to obtain the super-resolution model.

[0100] This involves using a super-resolution network to convert the target image into a target orthophoto; the resolution of the target orthophoto is greater than a set resolution value.

[0101] The third step is to build a classification model, which is also a detection model.

[0102] The detection model was set to a minimum batch size of 8, a training stride of 20, and a learning rate of 0.01. The final detection model was saved in *.MAT format. The learning rate refers to the magnitude of parameter updates during training of the shared neural network. The detection model included: CNN convolutional layers, max-pooling layers, and averpooling layers. The CNN convolutional layers used convolutional vector operations. Figure 3 To detect the parameters corresponding to each level of the model. Figure 3 The FC layer in this context refers to a fully connected layer, which can be represented as a projection process.

[0103] Step 4: Detecting virus-carrying strains using the sliding window method

[0104] This project uses MATLAB programming to process a target orthophoto. The positions of two points at the ends of grape rows in the orthophoto are obtained through a human-computer interaction. Based on the typical vine spacing of 1 to 2 meters for wine grapes in a vineyard, a sliding window size of 96*96 is set. The trajectory of the sliding window on the orthophoto is a straight line determined by the positions of the two ends of the grape rows.

[0105] The detection model from step 3, namely the SE-Resnet classification model, is invoked to detect the target window image acquired by the sliding window. The center point position of the sliding window is saved according to the classification results. Based on the geographic information of the orthophoto, the image position is converted into geographic coordinates and written into an SHP file to visualize the distribution of vineyard leafroll virus-carrying strains and the absence of vines.

[0106] In practical applications, steps 1-4 mentioned above do not necessarily have to be performed in this order. As long as each step is performed in accordance with the logical relationship to achieve the final detection objective, it is acceptable.

[0107] By applying UAV remote sensing technology to rapidly investigate the canopy growth of vineyards, and combining computer vision and deep learning algorithms, the model achieved precise location of vines infected with grape leafroll disease and accurate diagnosis of disease severity. It also detected missing vines due to leafroll and displayed the diagnostic results in the vineyard's geographic information system, visualizing the prevalence of grape leafroll. The accuracy of the grape canopy diagnostic model on the test set was 0.92. This result can provide a reference for refined management decisions in vineyards.

[0108] Example 2

[0109] In order to perform the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a diseased plant diagnosis and missing plant detection system is provided below. The system includes: an image processing module 1, a target window image acquisition module 2, and a detection information determination module 3.

[0110] Image processing module 1 is used to acquire target images of plants in the target area and generate target orthophotos based on the target images.

[0111] The target window image acquisition module 2 is used to divide the target orthophoto into multiple target window images by setting a sliding window.

[0112] The detection information determination module 3 is used to input multiple target window images into the detection model to obtain the detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease level of the plant.

[0113] When the disease severity level is zero, it indicates a lack of plants; the detection model is established using machine learning methods.

[0114] Example 3

[0115] This embodiment provides an electronic device, including a memory and a processor. The memory is used to store computer programs, and the processor runs the computer programs to enable the electronic device to perform the diseased plant diagnosis and missing plant detection method of Embodiment 1.

[0116] Alternatively, the aforementioned electronic device may be a server.

[0117] In addition, this embodiment of the invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the diseased plant diagnosis and missing plant detection method of Embodiment 1.

[0118] The advantages of the detection model and super-resolution network used in this invention are:

[0119] (1) The detection model is constructed using a deep learning network structure from machine learning, and its advantages are:

[0120] Compared to other deep learning network architectures, ResNet-50 achieved an accuracy of 0.66 on the test set, GoogLeNet achieved an accuracy of 0.52, and GoogLeNet is a novel deep learning architecture. Mobile V2 achieved an accuracy of 0.58 on the test set, the lightweight network ShuffleNet achieved an accuracy of 0.60, NasNet achieved an accuracy of 0.62, Place365-InceptionV3 achieved an accuracy of 0.52, VGG-16 achieved an accuracy of 0.52, and ResNet-18 achieved an accuracy of 0.64.

[0121] The classification accuracy was improved by adding the SE module. The SE module consists of two parts: compression and excitation. Compression (Squeeze) is achieved by performing global average pooling on the feature map layer to obtain the global compressed feature amount of the current feature map. Excitation (Excitation) is achieved by obtaining the weight of each channel in the feature map through a two-layer fully connected bottleneck structure (hourglass structure), and the weighted feature map is used as the input of the next layer of the network.

[0122] (2) Advantages of super-resolution networks, i.e., VDSR super-resolution methods

[0123] Peak signal-to-noise ratio (PSNR) of each image relative to a reference image and structural similarity index (SSIM) of each image are used as quantitative metrics to evaluate this super-resolution method. SSIM evaluates the visual impact of three features of an image on the reference image: brightness, contrast, and structure. The closer the SSIM value is to 1, the better the similarity between the test image and the reference image. A higher PSNR value generally indicates better image quality.

[0124] The PNSR and SSIM were evaluated by setting scaling factors for three gradients. The results show that, across all three gradients, VDSR outperforms the traditional bicubic interpolation method in obtaining super-resolution images of low-resolution images. When the scaling factor is 2, the PNSR using VDSR is 40.3710 and the SSIM is 0.9909; while the PSNR of the bicubic interpolation method is 38.5543 and the SSIM is 0.9866. When the scaling factor is 3, the PNSR using VDSR is 33.6644 and the SSIM is 0.9670; while the PSNR of the bicubic interpolation method is 33.4460 and the SSIM is 0.9643. When the scaling factor is 4, the PNSR using VDSR is 33.7361 and the SSIM is 0.9610; while the PSNR of the bicubic interpolation method is 32.7132 and the SSIM is 0.9542.

[0125] Therefore, the present invention can quickly and accurately detect plants.

[0126] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0127] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for diagnosing diseased plants and detecting missing plants, characterized in that, The method includes: Acquire target images of plants within the target area, and generate target orthophotos based on the target images; The target orthophoto image is divided into multiple target window images by using a set sliding window; Multiple target window images are input into the detection model to obtain detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease severity of the plant; When the disease severity level is zero, it indicates a lack of plants; The detection model is established using machine learning methods; The size of the sliding window is set according to the spacing between plants in the park; Generating a target orthophoto image from the target image specifically includes: After the target image undergoes geometric correction, feature point matching, image stitching, and georegistration, an orthophoto is generated. A super-resolution network is used to recover high-frequency information from low-frequency information in an orthophoto to obtain a target orthophoto; the resolution of the target orthophoto is greater than a set resolution value; the super-resolution network is determined using machine learning methods, with mean square error as the loss function to obtain the optimal peak signal-to-noise ratio.

2. The method for diagnosing diseased plants and detecting missing plants according to claim 1, characterized in that, The method of dividing the target orthophoto image into multiple target window images using a set sliding window specifically includes: The set sliding window is moved on the target orthophoto according to the set motion trajectory to obtain multiple target window images; the set motion trajectory is determined by the starting position and the ending position of the target orthophoto.

3. The method for diagnosing diseased plants and detecting missing plants according to claim 1, characterized in that, The method for determining the detection model is as follows: Obtain training data; The training data was obtained by augmenting the original training data using data augmentation methods. The original training data includes: original orthophotos of plants in different disease severity levels at each sampling point area within the experimental area. Multiple training window images are obtained by moving a set sliding window over the orthophoto of the training data; The multiple training window images are divided into a training set and a validation set; Construct a shared neural network; The training set is input into the shared neural network, and the parameters in the shared neural network are trained with the goal of minimizing the error, so as to obtain the trained shared neural network. The parameters of the trained shared neural network are adjusted using the validation set to obtain the detection model.

4. The method for diagnosing diseased plants and detecting missing plants according to claim 1, characterized in that, The method for determining the super-resolution network is as follows: Acquire training images; the training images include original images of plants in each sampling point area within the test area; Build a training network; The training image is input into the training network, and the peak signal-to-noise ratio of the training network is adjusted with the goal of minimizing the loss, to obtain the adjusted training network. The adjusted training network was determined to be a super-resolution network.

5. The method for diagnosing diseased plants and detecting missing plants according to claim 1, characterized in that, The method further includes: The detection information is classified and saved according to the severity of the disease.

6. A system for diagnosing diseased plants and detecting missing plants, characterized in that, The system includes: The image processing module is used to acquire target images of plants in the target area and generate target orthophotos based on the target images. The target window image acquisition module is used to divide the target orthophoto into multiple target window images using a set sliding window; The detection information determination module is used to input multiple target window images into the detection model to obtain detection information corresponding to each target window image; the detection information includes: the geographical location of the plant and the disease level of the plant; When the disease severity level is zero, it indicates a lack of plants; The detection model is established using machine learning methods; The size of the sliding window is set according to the spacing between plants in the park; Generating a target orthophoto image from the target image specifically includes: After the target image undergoes geometric correction, feature point matching, image stitching, and georegistration, an orthophoto is generated. A super-resolution network is used to recover high-frequency information from low-frequency information in an orthophoto to obtain a target orthophoto; the resolution of the target orthophoto is greater than a set resolution value; the super-resolution network is determined using machine learning methods, with mean square error as the loss function to obtain the optimal peak signal-to-noise ratio.

7. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the diseased plant diagnosis and missing plant detection method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the disease diagnosis and missing plant detection method as described in any one of claims 1 to 5.

Images