Rice disease identification method and system based on aerial images of planting areas

Abstract

The present invention relates to the field of rice disease identification. More specifically, the present invention relates to a method and system for rice disease identification based on aerial images of planting areas. The method includes: obtaining a plurality of preprocessed rice area images according to a preset path and constructing a rice image sequence; using the watershed algorithm to segment any rice area image to obtain a plurality of sub-regions, and constructing a multi-dimensional feature vector of the sub-regions; calculating the distance between any two sub-regions according to the feature vectors of the sub-regions, clustering the sub-regions according to the distance to obtain clustering clusters, and calculating the disease degree of the rice area image according to the clustering clusters, where the clustering clusters include normal clustering clusters and disease clustering clusters; traversing to obtain the disease degree of each rice area image in the rice image sequence to calculate the disease hazard degree of the planting area, and completing the disease identification. Through the technical solution of the present invention, the accuracy of the rice disease identification result can be improved.

Description

Technical Field

[0001] The present invention relates to the field of rice disease identification. More specifically, the present invention relates to a method and system for identifying rice diseases based on aerial images of planting areas. Background Art

[0002] As one of the important food crops globally, the growth, development, yield, and quality of rice are susceptible to the influence of pests and diseases. Common rice diseases such as rice blast, bacterial blight, brown spot, and false smut can all lead to reduced yield and quality, and in severe cases, may even cause large-scale production reduction. Traditional disease identification methods usually rely on manual inspections, which are not only time-consuming and laborious but also prone to missed or misdiagnosed cases. Therefore, how to use modern technical means to improve the efficiency and accuracy of rice disease identification has become an important topic in current agricultural research and practice.

[0003] Remote sensing technology, especially aerial images based on unmanned aerial vehicles (UAVs) or satellites, has been frequently applied in the agricultural field, mainly for pest and disease monitoring, crop growth status assessment, and soil quality analysis. Through aerial photography technology, it is possible to quickly and comprehensively monitor large areas of farmland and obtain important information about crop growth, pest and disease distribution, and soil conditions. This technology provides efficient data support for agricultural management, helping farmers and agricultural experts better grasp the real-time situation of farmland and make scientific decisions.

[0004] The Chinese patent application document with the publication number CN108563979A discloses a method for identifying the severity of rice blast based on aerial images of farmland. This application document uses the maximum inter-class variance method to segment the combined color component image based on color features to obtain rice disease spots. However, there will be cases where the RGB values of rice diseases are close to those of rice. Using only the combined color image component method to segment the image will result in inaccurate segmentation results. Summary of the Invention

[0005] To solve the problem of inaccurate segmentation results, the present invention proposes a method and system for identifying rice diseases based on aerial images of planting areas.

[0006] First aspect, the present invention discloses a rice disease recognition method based on aerial images of planting areas, including: obtaining a number of preprocessed rice area images according to a preset path and constructing a rice image sequence; using the watershed algorithm to segment any rice area image to obtain a number of sub-regions, constructing a multi-dimensional feature vector for the sub-regions, where one dimension in the feature vector corresponds to one water level line; calculating the distance between any two sub-regions according to the feature vectors of the sub-regions, clustering the sub-regions according to the distance to obtain clustering clusters, and calculating the disease degree of the rice area image according to the clustering clusters, where the clustering clusters include normal clustering clusters and disease clustering clusters; traversing to obtain the disease degree of each rice area image in the rice image sequence to calculate the disease hazard degree of the planting area and complete disease recognition; where the disease degree satisfies the relational expression:

[0007] , represents the disease degree, represents the area of the disease clustering cluster of, represents the area of the normal clustering cluster, represents the disease clustering cluster the proportion of the number of sub-regions in the total number of sub-regions, represents the normalization function, represents the logarithmic function.

[0008] By calculating the ratio of the area of the disease clustering cluster to the area of the normal area, the disease degree relational expression can reflect the comparison between the overall disease range of the rice and the healthy area, and intuitively reflect the diffusion degree of the disease. Considering the aggregation degree and chaos degree of different disease areas by using the entropy value, the larger the entropy value, the more types of diseases and the greater the disease degree.

[0009] Preferably, the preprocessing includes: obtaining the planting area image according to a preset path, and using the MTCNN model to extract the rice area image in the planting area image; using histogram equalization and contrast stretching to enhance the contrast and brightness of the rice area image, and using Gaussian filtering to denoise the rice area image.

[0010] The MTCNN model first accurately extracts the rice area in the planting area image, ensuring that subsequent processing focuses on the key parts, thus reducing the interference of irrelevant areas. Then, by using histogram equalization and contrast stretching techniques, the brightness and contrast of the rice area image are enhanced, making the details clearer. Especially in the case of uneven illumination or low contrast, it can better highlight the texture and disease characteristics of the rice. At the same time, Gaussian filtering is used to remove the noise in the image, reducing the impact of environmental interference on image analysis, so that subsequent disease recognition and analysis are more accurate.

[0011] Preferably, the extraction of the rice region image from the planting area image further includes: thresholding the planting area image using the G-channel information to obtain the rice region image.

[0012] The leaves of rice usually have a relatively high reflectance in the green band. Compared with other channels, the G-channel can better highlight the characteristics of rice vegetation. By extracting the G-channel image, the distribution of the rice region can be more accurately reflected.

[0013] Preferably, the feature vector is composed of the area of the sub-region and the number of edge pixel points of the sub-region together or is composed of the number of edge pixel points of the sub-region alone.

[0014] Preferably, the distance includes: calculating the distance between any two sub-regions using dynamic time warping according to the feature vectors of the sub-regions.

[0015] Preferably, the distance also satisfies the relational expression:

[0016] , represents the distance between sub-region and sub-region , where , represents the covariance between the feature vector of sub-region and the feature vector of sub-region , represents the standard deviation of the feature vector of sub-region , represents the standard deviation of the feature vector of sub-region , represents the exponential function.

[0017] The covariance measures the linear relationship between the feature vectors of two sub-regions, while the standard deviation characterizes the degree of dispersion of the features of each sub-region, further adjusting the scale of the similarity measure. This distance metric can not only reveal the mutual relationship between different disease regions or healthy regions within the rice region, but also effectively distinguish the distribution pattern and evolution trend of diseases during the image analysis process.

[0018] Preferably, the calculation of the disease damage degree of the planting area includes: setting the rice region images with a disease degree not less than the preset disease threshold in the rice image sequence to 1, setting the rice region images with a disease degree less than the preset disease threshold in the rice image sequence to 0, converting the rice image sequence into a binary sequence, and counting the number of consecutive occurrences of 1; the disease damage degree satisfies the relational expression:

[0019] , represents the disease damage degree, Represents the average disease severity of the rice area images in the rice image sequence. Represents the variance of the number of consecutive occurrences of 1. Represents the normalization function.

[0020] The binary sequence reflects the persistence and spread trend of the disease. By calculating the average disease severity and the variance of the number of consecutive severe areas, the distribution and volatility of the disease condition are comprehensively evaluated.

[0021] In a second aspect, the present invention discloses a rice disease recognition system based on aerial images of the planting area, including: a processor; and a memory storing computer instructions that, when run by the processor, cause the system to execute the above-mentioned rice disease recognition method based on aerial images of the planting area.

[0022] Advantages of the present invention:

[0023] The present invention first optimizes the image quality through preprocessing techniques, extracts the rice area and enhances its contrast and brightness, making the disease manifestation more obvious. Then, the watershed algorithm is used to finely segment the rice image, the rice area is refined into multiple sub-regions, its multi-dimensional feature vectors are calculated, and through dynamic time warping and distance calculation methods, the differences between the disease clustering clusters and the normal areas are identified, so as to accurately evaluate the disease severity of each area. Finally, by statistically analyzing the disease severity and the change trend in the image sequence, the disease hazard degree of the entire planting area can be quantified.

[0024] The present invention can efficiently and accurately identify diseases in large-scale rice planting areas, avoid the lag of manual detection, and effectively improve the efficiency of pest and disease management in rice planting. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] By referring to the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0026] Figure 1 is a flowchart of the rice disease recognition method based on aerial images of the planting area according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] It should be understood that when terms such as "first", "second", etc. are used in the claims, specification and drawings of the present invention, they are only used to distinguish different objects, rather than to describe a specific order. The terms "comprising" and "including" used in the specification and claims of the present invention indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0029] The present invention provides a method for identifying rice disease conditions based on aerial images of planting areas. As Figure 1 shown, the method for identifying rice disease conditions based on aerial images of planting areas includes steps S1 - S4, which are specifically described below.

[0030] S1, obtain a number of pre - processed rice area images according to a preset path and construct a rice image sequence.

[0031] In one embodiment, a high - resolution image of the planting area is obtained by a drone according to a preset flight path, and the MTCNN (Multi - task Cascaded Convolutional Networks) model is used to process the image to automatically identify and extract multiple rice area images in the image.

[0032] For each rice area image, first, the histogram equalization technique is applied to improve the brightness distribution of the image, thereby enhancing the detail performance of the image; subsequently, the contrast stretching method is used to further enhance the contrast of the image, making the features of the rice area more prominent. Then, in order to eliminate the noise in the rice area image, the Gaussian filtering algorithm is used to smooth the image, reducing the noise interference in the details and improving the image quality. After these pre - processing steps, high - quality and clear rice area images are obtained. Finally, according to the preset path of the drone flight, these processed rice area images are organized into a rice image sequence in chronological order or spatial order.

[0033] In one embodiment, extracting the rice area images from the planting area image further includes: performing threshold segmentation on the planting area image using the G - channel information to obtain the rice area image.

[0034] It should be noted that first, the G - channel (i.e., the green channel) of the planting area image is extracted because the leaves of rice usually have a higher reflectance in the green band. Compared with other channels (such as the red channel and the blue channel), the G - channel can better highlight the characteristics of rice vegetation. By extracting the G - channel image, the distribution of the rice area can be more accurately reflected.

[0035] Next, select an appropriate threshold. Usually, the threshold can be set manually or an automated thresholding method can be used for segmentation. The purpose of threshold segmentation is to distinguish the rice area in the image from the background area. Specifically, in the G-channel image, the rice area usually has higher pixel values, while the background area (such as soil, open space, etc.) has lower pixel values. By setting a threshold, the area with pixel values higher than the threshold is marked as the rice area, and the area with pixel values lower than the threshold is marked as the background area.

[0036] After completing the threshold segmentation, a binary image of the rice area can be obtained, where the rice area is usually represented by white and the background area is represented by black. To further refine the extraction of the rice area, morphological operations can be used to remove noise, smooth the boundaries, and improve the accuracy of the segmentation result.

[0037] S2. Use the watershed algorithm to segment any rice area image to obtain several sub-regions, and construct multi-dimensional feature vectors for the sub-regions.

[0038] It should be noted that when analyzing the rice area, first, precise segmentation of the rice area is carried out through image processing techniques to extract key parts such as leaves, because the diseases of rice usually manifest as abnormalities in the leaves. The diseases of rice often lead to visible changes on the leaves, such as holes bitten by insects, spots, and browning on the leaves. These diseased areas often have different color, texture, or morphological characteristics from healthy leaves. Therefore, image segmentation can not only help identify the healthy areas of rice but also accurately distinguish the damaged leaf areas. After obtaining different regions through segmentation, the size, distribution, and type of these diseased areas can be further analyzed to evaluate the severity of the rice disease.

[0039] The watershed algorithm divides the image into multiple different sub-regions by simulating the process of water flow gradually eroding in the terrain, and each sub-region represents an area with similar visual characteristics. On this basis, to more comprehensively analyze the characteristics of the rice area, multi-dimensional feature vectors can be constructed for each sub-region, where each dimension corresponds to the "water level line" in the image - that is, in the watershed algorithm, adjacent regions are gradually merged at different stages of the rising water level, and finally independent segmentation areas are formed. Each dimension reflects different levels of image features, such as color, texture, shape, edge information, etc. Specifically, these "water level lines" can help capture different segmentation levels in the image, thereby expressing various features of the region as numerical vectors.

[0040] In one embodiment, the feature vector is jointly composed of the area of the sub-region and the number of edge pixel points of the sub-region or is solely composed of the number of edge pixel points of the sub-region.

[0041] When it is composed solely of the number of edge pixels of the sub-region, the feature vector is a 1-by- matrix, where represents the number of dimensions, and the value of each dimension in the feature vector represents the number of edge pixels of the sub-region when the water level line is .

[0042] When it is composed of the sub-region area and the number of edge pixels of the sub-region together, the feature vector is a 2-by- matrix, where represents the number of dimensions, and the value of each dimension in the feature vector represents the number of edge pixels of the sub-region and the sub-region area when the water level line is .

[0043] S3. Calculate the distance between any two sub-regions according to the feature vectors of the sub-regions, obtain clustering clusters by clustering the sub-regions according to the distance, and calculate the disease severity of the rice region image according to the clustering clusters.

[0044] In one embodiment, the distance between any two sub-regions is calculated using dynamic time warping according to the feature vectors of the sub-regions.

[0045] Cluster the sub-regions according to the distance to obtain clustering clusters. The clustering clusters include normal clustering clusters and disease clustering clusters. To distinguish the normal clustering clusters, a clustering cluster all of which are disease-free can be added as a template clustering cluster, and the pixel points of each clustering cluster and the template clustering cluster are calculated to distinguish the normal clustering clusters.

[0046] Calculate the disease severity of the rice region image according to the clustering clusters. The disease severity satisfies the relational expression:

[0047] , represents the disease severity, represents the area of the disease clustering cluster , represents the area of the normal clustering cluster, represents the proportion of the number of sub-regions in the disease clustering cluster to the total number of sub-regions, represents the normalization function, represents the logarithmic function.

[0048] By calculating the ratio of the area of the disease clustering cluster to the area of the normal region (leaf), the range of the overall disease of the rice can be reflected in comparison with the healthy region, and the spread degree of the disease can be intuitively reflected. The whole is the entropy value. Considering the aggregation degree and chaos degree of different disease regions, the larger the entropy value, the more types of diseases and the greater the disease severity.

[0049] In one embodiment, the distance also satisfies the relational expression:

[0050] , represents the distance between sub-regions and sub-region where , represents the covariance between the eigenvectors of sub-region and the eigenvectors of sub-region . represents the standard deviation of the eigenvectors of sub-region . represents the standard deviation of the eigenvectors of sub-region . represents the exponential function.

[0051] Covariance measures the linear relationship between the eigenvectors of two sub-regions, while the standard deviation characterizes the degree of dispersion of the features of each sub-region, further adjusting the scale of the similarity measure. By using the exponential function to transform these quantities, sub-regions with higher similarity have shorter distances, while sub-regions with lower similarity have longer distances, thus providing a smooth and flexible similarity measurement method. This distance measurement method can not only reveal the relationships between different disease regions or healthy regions within the rice region, but also effectively distinguish the distribution patterns and evolution trends of diseases during the image analysis process.

[0052] S4. Traverse to obtain the disease severity of each rice region image in the rice image sequence to calculate the disease damage degree of the planting area and complete the disease identification.

[0053] It should be noted that the disease damage degree of the planting area is not only related to the single disease severity in each image, but also closely related to the spatial distribution of the disease severity. When the areas with relatively severe disease severity are concentrated in some adjacent rice region images, it usually indicates that the disease is limited to some local areas of the planting area, and the spread and diffusion of the disease have not formed a large-scale impact. At this time, the control measures may be concentrated on these specific areas to effectively control the development of the disease. However, if the areas with severe disease appear in the rice region images at different position segments, it means that the disease has occurred not only in some local areas but also has spread to multiple places in the planting area, indicating that the disease is spreading and the control difficulty has increased significantly. At this time, the disease in the entire planting area has begun to show a spreading trend, meaning that the disease spreads faster and may cross multiple regions. The prevention and control measures need to be more comprehensive and timely to avoid further spread of the disease and cause greater losses.

[0054] In one embodiment, the rice area images in the rice map sequence with a disease severity not less than a preset disease threshold are set to 1, the rice area images in the rice map sequence with a disease severity less than the preset disease threshold are set to 0, the rice map sequence is converted into a binary sequence, and the number of consecutive occurrences of 1 is counted.

[0055] The disease hazard degree satisfies the relational expression: , represents the disease hazard degree, represents the average disease severity of the rice area images in the rice map sequence, represents the variance of the number of consecutive occurrences of 1, represents the normalization function.

[0056] Exemplarily, the binary sequence counted is , then the number of consecutive occurrences of 1 is 2, 1, and 3. Then the variance in the formula is the variance of 2, 1, and 3.

[0057] Compare the disease hazard degree with the danger threshold. When the disease hazard degree is not less than the danger threshold, generate and send an alarm signal to complete the disease identification.

[0058] The embodiment of the present invention also discloses a rice disease identification system based on aerial images of the planting area, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the rice disease identification method based on aerial images of the planting area according to the present invention is implemented.

[0059] The above system also includes other components well-known to those skilled in the art such as a communication bus and a communication interface, and their settings and functions are known in the art, so they will not be described in detail here.

[0060] In the present invention, the aforementioned memory may be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, the computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application program, module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

[0061] Although this specification has shown and described multiple embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many variations, changes, and alternative ways will occur to those skilled in the art without departing from the spirit and scope of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

[0062] The above are all preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A rice disease identification method based on aerial images of planting areas, characterized in that: include: Acquire a number of pre-processed rice area images according to a preset path, and construct a rice image sequence; The watershed algorithm is used to segment any rice area image to obtain several sub-areas, and a multi-dimensional feature vector of the sub-area is constructed, where one dimension in the feature vector corresponds to one water level line. The distance between any two sub-regions is calculated according to the feature vectors of the sub-regions, the sub-regions are clustered according to the distance to obtain clusters, and the disease severity of the rice region image is calculated according to the clusters, and the clusters include normal clusters and disease clusters; The disease severity of each rice region image in the rice image sequence is obtained by traversing to calculate the disease severity of the planting area and complete disease identification; the disease severity satisfies the relationship: , Indicates the severity of the disease, Indicates disease clusters The area of represents the area of ​​a normal cluster, Indicates disease clusters The ratio of the number of neutron regions to the total number of sub-regions, represents the normalization function, represents the logarithmic function; The calculation of the degree of disease damage in the planting area includes: The rice region images whose disease severity is not less than the preset disease severity threshold in the rice image sequence are set to 1, and the rice region images whose disease severity is less than the preset disease severity threshold in the rice image sequence are set to 0, and the rice image sequence is converted into a binary sequence, and the number of consecutive occurrences of 1 is counted; The degree of disease damage satisfies the relationship: , Indicates the severity of the disease. represents the mean disease severity of the rice area images in the rice image sequence, represents the variance of the number of consecutive occurrences of 1, Represents the normalization function.

2. The rice disease identification method based on aerial images of planting areas according to claim 1 is characterized in that: The pre-processing comprises: Obtain the planting area image according to the preset path, and use the MTCNN model to extract the rice area image in the planting area image; Histogram equalization and contrast stretching are used to enhance the contrast and brightness of rice area images, and Gaussian filtering is used to denoise the rice area images.

3. The rice disease identification method based on aerial images of planting areas according to claim 2 is characterized in that: The step of extracting the rice area image from the planting area image further comprises: The G channel information is used to perform threshold segmentation on the planting area image to obtain the rice area image.

4. The rice disease identification method based on aerial images of planting areas according to claim 1 is characterized in that: The feature vector is composed of the sub-region area and the number of edge pixels in the sub-region or is composed of the number of edge pixels in the sub-region alone.

5. The rice disease identification method based on aerial images of planting areas according to claim 1 is characterized in that: The distances include: The distance between any two sub-regions is calculated using dynamic time warping based on the feature vectors of the sub-regions.

6. The rice disease identification method based on aerial images of planting areas according to claim 1 is characterized in that: The distance also satisfies the relationship: , Indicates sub-area and subregions The distance , Indicates sub-area The eigenvector and subregion of The covariance of the eigenvectors of Indicates sub-area The standard deviation of the eigenvectors, Indicates sub-area The standard deviation of the eigenvectors, Represents an exponential function.

7. A rice disease identification system based on aerial images of planting areas, characterized in that: include: processor; as well as A memory storing computer instructions, wherein when the computer instructions are executed by the processor, the system executes the rice disease identification method based on aerial images of the planting area according to any one of claims 1 to 6.

Images