Method for identifying diseases of rice by using unmanned equipment, unmanned equipment and device

Abstract

The invention discloses a method for identifying diseases of rice by using unmanned equipment, the unmanned equipment and a device, the unmanned equipment comprises an unmanned aerial vehicle, a connecting mechanism and an image collecting device, and the image collecting device and the unmanned aerial vehicle are connected through the connecting mechanism and the distance between the image collecting device and the unmanned aerial vehicle is controlled through the connecting mechanism; comprising the steps that a task instruction is obtained and sent to a corresponding unmanned device, the task instruction is used for enabling an unmanned aerial vehicle to be in a hovering state when the unmanned aerial vehicle reaches a task area above rice, a connecting mechanism is controlled to be prolonged, and an image collecting device moves downwards and penetrates through a leaf layer of the rice till the image collecting device is close to the water surface; the image of the rice close to the water surface is obtained; and inputting the image into the trained disease identification neural network to obtain an identification result. According to the technology, the image collecting device is controlled to penetrate through the rice leaf layer till the position close to the water surface for image collection, detection of diseases on the upper portion of the rice water surface is achieved, and early detection of the diseases such as banded sclerotial blight is ensured.

Description

Technical Field

[0001] The present application relates to the field of disease identification, and more specifically, to a method, unmanned equipment and device for identifying rice diseases using unmanned equipment. Background Art

[0002] With the increasing use of drones, drones have also penetrated into agricultural applications. In the prior art, drones are often used to detect rice diseases.

[0003] However, in the process of rice cultivation, some diseases are prone to appear only on the upper part of the rice surface in the early stage, such as sheath blight. As the rice leaves gradually grow luxuriantly, it is difficult to find these diseases from the pictures above the rice taken by drones using existing technology, thus missing the best time to find the diseases.

[0004] Accordingly, a new disease identification solution is needed in this field to solve the above problems. Summary of the invention

[0005] In order to overcome the above defects, the present invention is proposed to provide a solution or at least partially solve the problem in the prior art that it is difficult to obtain images near the upper part of the rice water surface using a drone.

[0006] In a first aspect, the present invention provides a method for identifying rice diseases using unmanned equipment, wherein the unmanned equipment includes a drone, a connecting mechanism, and an image collection device, wherein the image collection device and the drone are connected and distance-controlled via the connecting mechanism, and the method includes: obtaining a task instruction, and sending the task instruction to the corresponding unmanned equipment, wherein the task instruction is used to control the drone to be in a hovering state when the drone reaches a task area above the rice, and to control the connecting mechanism to extend so that the image collection device moves downward, passes through a leaf layer of the rice until it is close to the water surface, so as to obtain an image of the rice close to the water surface; and inputs the image into a trained disease recognition neural network to obtain a recognition result.

[0007] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the method also includes: obtaining location information corresponding to the recognition result, wherein the location information is used to characterize the shooting location of the image; based on the location information, performing spatial distribution visualization processing on the recognition result to obtain a disease distribution map.

[0008] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the step of acquiring an image of rice near the water surface includes: acquiring N block images, wherein the N block images correspond to the same position information, the position information is used to characterize the position of the drone when acquiring the block images, and the N block images are used to characterize the images captured by the image acquisition device at different viewing angles, and there are repeated areas between two block images of adjacent viewing angles; the step of inputting the images into a trained disease recognition neural network to obtain a recognition result includes: acquiring N block images with the same position information; acquiring the N block images; The same-name points of the N block images are identified, and the N block images are registered based on the image name points to obtain the registered N block images, wherein the same-name points are used to characterize the pixel points representing the same actual position in different block images; the N block images after registration are input into the disease recognition neural network to obtain the recognition results of all the disease points in the N block images, wherein the recognition results include marking the disease position; the method further includes: judging whether the marks in different block images are repeated marks, and merging the repeated marks based on the judgment result to obtain a unique mark corresponding to a disease position. .

[0009] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, determining whether marks in different block images are repeated marks includes: determining whether marks in different block images are points of the same name, and if there are multiple marks that are points of the same name, then the multiple marks are repeated marks.

[0010] In a second aspect, a training method for a disease recognition neural network is provided, the method comprising: sending different task instructions to a drone fleet, wherein the drone fleet comprises a plurality of unmanned devices, any one of the task instructions corresponds to one of the unmanned devices, the unmanned devices comprising a drone, a connecting mechanism and an image collection device, wherein the image collection device and the drone are connected and distance-controlled via the connecting mechanism, wherein the task instruction is used to control the drone to be in a hovering state when the drone reaches a task area above rice, and to control the connecting mechanism to extend so that the image collection device moves downward, passes through a leaf layer of rice until close to the water surface, so as to obtain an image of the rice close to the water surface; annotating the image to obtain an image sample; preprocessing the image sample, and inputting the preprocessed image sample into a disease recognition neural network to obtain a recognition result, and training the disease recognition neural network using a back propagation algorithm based on the recognition result and the label of the image sample.

[0011] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the preprocessing includes: normalization, quality control, data augmentation and cleaning, feature recognition, feature importance recognition and feature selectivity enhancement of image samples; in training the disease recognition neural network, the methods used include: activation function, batch normalization function, random inactivation and full connection.

[0012] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the step of acquiring an image of rice near the water surface includes: acquiring N block images, wherein the N block images correspond to the same position information, the position information is used to characterize the position of the drone when acquiring the block images, and the N block images are used to characterize images captured by the image acquisition device at different viewing angles, and there are repeated areas between two block images of adjacent viewing angles; the step of inputting the preprocessed image samples into the disease recognition neural network to obtain the recognition result includes: acquiring N block image samples with the same preprocessed position information; acquiring the N block images; The method further comprises: determining the same-name points of the N block image samples, and performing image registration on the N block image samples based on the image name points to obtain the registered N block image samples, wherein the same-name points are used to characterize the pixel points representing the same actual position in different block image samples; inputting the registered N block image samples into the disease recognition neural network to obtain the recognition results of all the disease points in the N block image samples, wherein the recognition results include marking the disease positions; and the method further comprises: determining whether the marks in the different block image samples are repeated marks, and merging the repeated marks based on the determination result to obtain a unique mark corresponding to a disease position.

[0013] In a third aspect, an unmanned device is provided, which includes: a drone and an image collecting device, wherein the image collecting device is connected to the drone via a connecting mechanism, and the length of the connecting mechanism can be changed to adjust the distance between the drone and the image collecting device.

[0014] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the unmanned equipment also includes a floating device and a weight, wherein the floating device is used to support the image collecting device and make the image collecting device float on the water surface, the weight is located at the lower part of the floating device, and the upper limit of the buoyancy provided by the floating device is greater than the sum of the gravity of the image collecting device and the weight.

[0015] In a fourth aspect, a control device is provided, which includes a processor and a storage device, wherein the storage device is suitable for storing multiple computer programs, and the computer programs are suitable for being loaded and run by the processor to execute the method for identifying rice diseases using unmanned equipment as described in any of the technical solutions of the above-mentioned method for identifying rice diseases using unmanned equipment.

[0016] In a fifth aspect, a computer-readable storage medium is provided, wherein a plurality of computer programs are stored in the computer-readable storage medium, wherein the computer programs are suitable for being loaded and run by a processor to execute the method for identifying rice diseases using unmanned equipment as described in any one of the technical solutions of the method for identifying rice diseases using unmanned equipment.

[0017] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0018] In the technical solution for implementing the present invention, by configuring an image collection device connected to a drone in the unmanned equipment and controlling it to pass through the rice leaf layer until it is close to the water surface to collect images, the technology realizes the early detection of diseases above the water surface of rice, ensuring the early discovery of diseases such as sheath blight. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0020] In order to more clearly illustrate the embodiments of the present application 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, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 is a schematic diagram of the structure of an unmanned device according to an embodiment of the present invention;

[0022] Figure 2a This is a diagram of a drone hovering over a rice field, ready to begin imaging work;

[0023] Figure 2b This is a diagram of a drone hovering over a rice field and starting image processing;

[0024] Figure 3 It is a schematic diagram of an image collection device acquiring an image of a disease on the upper part of the water surface of rice;

[0025] Figure 4 is a schematic diagram of a floating device and a weight in one embodiment;

[0026] Figure 5 is a flow chart of main steps of a method for identifying rice diseases using unmanned equipment according to an embodiment of the present invention;

[0027] Figure 6 It is a schematic diagram of a blade layer;

[0028] Figure 7 It is a schematic diagram of a panoramic image and a block image;

[0029] Figure 8 This is a flow chart of the training process of a disease recognition neural network.

[0030] List of reference numerals:

[0031] 1. Drone; 2. Connecting mechanism; 3. Image collection device; 4. Floating device; 5. Load-bearing. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] Pitch angle: The vertical shooting angle of the camera, usually used to describe the change in the camera's viewing angle from top to bottom or from bottom to top.

[0035] In order to identify rice diseases, a large number of unmanned equipment (drone swarms) are used to carry out the task.

[0036] The unmanned equipment includes a drone 1 and an image collection device 3 .

[0037] In some embodiments, the image collecting device 3 is a camera capable of penetrating the leaf layer of rice.

[0038] like Figure 1 As shown, Figure 1 It is a schematic diagram of the structure of an unmanned device in one embodiment. The image collecting device 3 can pass through the leaf layer of rice with the assistance of other components. For example, the unmanned device also includes: a connecting mechanism 2. The image collecting device 3 is stably connected to the bottom of the drone 1 through the connecting mechanism 2 to achieve a reliable vertical suspension effect. The connecting mechanism 2 can be a load-bearing line or a connecting rod. The connecting mechanism 2 ensures that the image collecting device 3 can pass through the leaf layer when the drone 1 is located above the leaf layer, and obtain image information of the rice below the leaf layer.

[0039] In some embodiments, the image collection device 3 is a panoramic camera. The panoramic camera can capture a complete image of the surrounding environment from different angles through a surround lens and multiple image sensors, thereby achieving all-round monitoring of the area. In the identification of diseases in rice fields, the panoramic camera can quickly obtain a panoramic image under the leaves while the drone 1 is hovering, and capture the distribution of diseases, thereby providing more comprehensive data support for disease monitoring. In addition, the image collection device 3 can also be an infrared camera or a multi-spectral camera. Infrared cameras can capture the thermal radiation information of plants and are used to detect the health of crops. Diseased areas usually show different temperature characteristics due to transpiration and changes in leaf structure, and infrared imaging can quickly detect early diseases. Multi-spectral cameras can capture light in multiple bands (such as visible light and near infrared) and evaluate plant health by analyzing vegetation indices (such as NDVI). Healthy plants and plants affected by diseases reflect different light intensities under different spectra, which facilitates early identification of diseases.

[0040] The connecting mechanism 2 is used to connect the image collection device 3 and the drone 1. The connecting mechanism 2 can be of a fixed length or of a variable length. In this embodiment, the length of the connecting mechanism 2 can be varied, that is, the length of the connecting mechanism 2 can be adjusted to adjust the distance between the drone 1 and the image collection device 3. In some embodiments, the connecting mechanism 2 can be a telescopic rod or a device composed of a load-bearing rope and a retractable wire device. In the following examples, the example in which the connecting mechanism 2 includes a load-bearing rope and a retractable wire device is used for explanation.

[0041] The retractable wire device can be an electric mechanical device for controlling the lifting and lowering of the image collection device 3, and the image collection device 3 can be moved up and down when the drone 1 is hovering by winding and releasing the load-bearing wire. The retractable wire device accurately adjusts the height of the image collection device 3 to ensure that it can pass through the leaf layer of the crop and approach a specific position (such as the water surface) during the hovering of the drone 1 to obtain the best image acquisition angle and viewing angle. The retractable wire device works in coordination with the electric motor and the control system to provide a stable and adjustable lifting speed, ensuring the smoothness of the acquisition process and the stability of the image.

[0042] like Figure 2a and Figure 2b As shown. Figure 2a This is a schematic diagram showing that UAV 1 is hovering over the rice field but has not yet started imaging operations. Figure 2b FIG. 1 is a schematic diagram of UAV 1 hovering over a rice field and starting image processing. Figure 2a and Figure 2b It can be further known that the image collection device 3 can be retracted and extended by the connection mechanism 2, so that the image collection device 3 can be close to the water surface and close to the upper part of the water surface of the rice. In this embodiment, the upper part of the water surface of the rice refers to the rice part of the structure located at the water surface and above the water surface.

[0043] exist Figure 2a In the process, the drone 1 will first go through a long flight. During the flight of the drone 1, the safety of the drone 1 must be ensured. At this time, an overly long connection mechanism 2 is likely to hinder the normal flight of the drone 1. For example, when the exposed load-bearing line is too long, the exposed line is likely to collide with other objects during the flight of the drone 1, causing the drone 1 to be hung on various objects (such as tree branches). Therefore, for safety reasons, during the flight of the drone 1, the connection mechanism 2 is controlled to shorten (the exposed load-bearing line is shortened) so that the drone 1 and the image collection device 3 are almost close together to reduce the risk of bombing. After the drone 1 flies above the field, it is necessary to obtain an image of the upper part of the rice water surface. The upper part of the rice water surface is located below the rice leaves and near the water surface. However, due to the obstruction of the rice leaves, it is impossible for the drone 1 to personally reach down to the upper part of the rice water surface to take images, so the probability of collision between the propeller and the blades is too high. Therefore, in this embodiment, if Figure 2b At this time, the drone 1 starts the suspension mode and continuously extends the connection mechanism 2 (by operating the retractable wire device to continuously release the load-bearing wire), so that the length of the connection mechanism 2 becomes longer and longer, so that the image collection device 3 is stably placed at a preset height under the premise that the drone 1 is in a safe position to ensure that the image of the upper part of the rice water surface is obtained at the best viewing angle. Figure 3 As shown, Figure 3It is a schematic diagram of an image collection device 3 acquiring disease images on the upper part of the rice water surface. It can be seen that when the image collection device 3 is at the water surface, it can bypass the shielding of the rice leaf layer and acquire the diseased spot image on the upper part of the rice water surface.

[0044] In the above example, the drone 1 completes the above operation according to the task instruction, and similarly, the image collection device 3 also acquires images according to the task instruction. In addition to the physical connection between the image collection device 3 and the drone 1, there is also a connection on the communication line. In some embodiments, the unmanned device also includes a control transmission line, wherein the drone 1 is connected to the image collection device 3 for communication via the control transmission line. In addition, the drone 1 can also connect the two for communication via a wireless connection. In one embodiment, a control transmission line is used for communication connection, and the transmission of instructions using the control transmission line can help reduce the power consumption of the drone 1. Specifically, when wireless transmission is performed, the drone 1 and the image transmission device must be installed with an additional wireless module, wherein the wireless module must have receiving and transmitting functions, that is, there must be receiving and transmitting submodules. Such a complex wireless module will increase the power consumption of the drone 1, which is not conducive to the endurance of the unmanned device, and will increase the manufacturing cost of the overall unmanned device.

[0045] Furthermore, the image collection device 3 is also equipped with a weight 5. By adding a weight 5 of appropriate weight to the bottom of the image collection device 3, it is ensured that the center of gravity of the unmanned equipment moves downward in the suspended state, thereby improving the wind resistance of the unmanned equipment. In addition, the weight 5 will make the image collection device 3 more resistant to obstruction. Specifically, the weight 5 can prevent the image collection device 3 from being unable to fall simply because of the obstruction of a small number of blades. By applying additional gravity to the blades, the blades are disturbed by external forces and tilted or deformed in other ways, so that the image collection device 3 can continue to move downward.

[0046] Further, such as Figure 4 As shown, Figure 4 It is a schematic diagram of a floating device 4 and a weight 5 in an embodiment. The image collecting device 3 is also equipped with a floating device 4, which is used to support the image collecting device 3 and make the image collecting device 3 float on the water surface. When the floating device 4 and the weight 5 exist together, the weight 5 is located below the floating device 4. At this time, the upper limit of the buoyancy provided by the floating device 4 is greater than the sum of the gravity of the image collecting device 3 and the weight 5. When the floating device 4 is working (when the floating device 4 touches the water surface to provide buoyancy), the weight 5 can also provide the function of increasing the stability of the floating device 4.

[0047] like Figure 4As shown, the floating device 4 is in a "bowl" shape to ensure that the image collection device 3 floats on the water surface. The floating device 4 is usually made of a lightweight and water-resistant material, and the floating device 4 enables the image collection device 3 to remain stable on the water surface. By using the floating device 4, it is ensured that the image acquisition device can stay stably on the water surface, thereby achieving more accurate image data acquisition.

[0048] When identifying rice diseases, the above unmanned equipment is used to complete the disease identification, as follows:

[0049] See attached Figure 5 , Figure 5 FIG. 1 is a flow chart showing the main steps of a method for identifying rice diseases using unmanned equipment according to an embodiment of the present invention. Figure 5 As shown, the method for identifying rice diseases using unmanned equipment in the embodiment of the present invention mainly includes the following steps S10 to S30.

[0050] Step S10: In response to the task instruction, the unmanned equipment performs image acquisition according to the task instruction.

[0051] The task instruction refers to a set of instructions preset by the system or input by the user in real time for controlling the unmanned equipment to perform the rice disease identification task. In some embodiments, the user may only issue an original instruction, and then the system analyzes the original instruction to obtain specific task instructions for each unmanned equipment.

[0052] For example, there are three rice fields, namely area A, area B and area C. The user may issue an original instruction in the form of a UI interface, that is, to identify diseases in the above three areas, and then select the number of unmanned devices that can be dispatched, that is, the number of unmanned devices in the drone fleet, such as 50 unmanned devices in the drone fleet. After that, the system can spontaneously assign tasks to the 50 unmanned devices, and fuzzy algorithms and other methods can be used here to assign tasks. After that, the task instructions corresponding to each unmanned device are obtained, and different task instructions are sent to the corresponding different unmanned devices.

[0053] In one implementation, the mission instruction includes at least parameters such as acquisition time, flight path, image acquisition location, and flight altitude.

[0054] The mission instruction is used to control the image collection device 3 to pass through the leaf layer of the rice until it is close to the water surface when the drone 1 reaches the mission area above the rice, so as to obtain the image of the rice close to the water surface.

[0055] During the descent of the image collection device 3, the drone 1 is in a hovering state in order to maintain the stability of the image collection device 3. The image collection device 3 will continue to descend and pass through the rice leaf layer. Figure 6 As shown, Figure 6 This is a schematic diagram of the leaf layer. The leaf layer refers to the overall structure of staggered leaves formed during the growth of rice, covering the rice stems and the upper part of the water surface. The leaf layer is densely covered in the farmland, which can protect the upper part of the water surface of the crop from direct sunlight and maintain humidity, while providing conditions for the concealment of pests and diseases. The leaf layer is not fixed. As the rice grows, the leaf layer becomes gradually denser, covering the rice stems and the upper part of the water surface.

[0056] like Figure 6 As shown, below the area where the leaves of the two rice plants intersect, there is a blank area on the water surface, and this blank area is located between the two rice plants. Then the image collection device 3 will descend to this position to collect images.

[0057] Specifically, when the drone 1 flies to the designated mission area, the drone 1 changes from a cruising state to a hovering state. And through the operation of the retracting and releasing device, the load-bearing line is released, and then the image collection device 3 gradually descends under the influence of gravity. After that, the image collection device 3 will pass through the blade layer and then continue to approach the water surface. Until it reaches a predetermined height, the connecting mechanism 2 is fixed at this time. Alternatively, when the floating device 4 touches the water surface, the floating device 4 functions and supports the image collection device 3 on the water surface.

[0058] The image collection device 3 will obtain images of the rice near the water surface. These positions below the leaf layer belong to the stem of the rice and the upper part of the water surface. In this technology, the useful position is the upper part of the rice on the water surface. Acquiring images of the upper part of the rice on the water surface can effectively capture early symptoms, such as symptom information of sheath blight. Sheath blight or similar diseases often occur in the upper part of the rice near the water surface in the early stages. This step achieves accurate collection of images of the disease-occurring site by inserting the image collection device 3 deep into the rice leaf layer and close to the water surface, which is helpful for the subsequent training of disease recognition models and the identification of rice diseases.

[0059] In the process of acquiring images, the image collection device 3 acquires images from multiple angles, that is, usually acquires images containing multiple upper portions of the rice water surface. By acquiring multiple images of the upper portions of the rice water surface in one acquisition, the number of samples is increased, and the probability of discovering rice diseases is also increased. By acquiring images from multiple angles, data coverage is maximized, which helps to improve the reliability and effectiveness of sheath blight detection.

[0060] After the image acquisition is completed, the drone 1 will pull up the image collection device 3 through the retractable wire device. Then, it will fly to the next task area according to the requirements of the task instruction, and continue to perform the image acquisition operation until the entire task instruction is completed. The number of unmanned equipment varies depending on the area of ​​the rice field. The number of task areas that the unmanned equipment is going to collect in the task instruction is different. It can be imagined that if the area of ​​the rice field is large, but the number of unmanned equipment is small, then one unmanned equipment in the task instruction may involve multiple task areas, that is, one unmanned equipment is responsible for the collection of images of multiple task areas. But on the contrary, if the area of ​​the rice field is small, but the number of unmanned equipment is sufficient, then one unmanned equipment in the task instruction may be responsible for only a small number of task areas, or even only one task area.

[0061] By executing mission instructions, unmanned equipment can achieve comprehensive, efficient and orderly image acquisition of rice fields, providing comprehensive and reliable image data support for subsequent image processing and disease identification.

[0062] Step S20: input the image into a trained disease recognition neural network to obtain a recognition result.

[0063] In this embodiment, the trained disease recognition neural network can recognize the image and evaluate the degree of disease of the rice in the image.

[0064] In some embodiments, the image input to the disease recognition neural network is a block image. The block images will correspond to the panoramic image after being combined in order. The block images are generally formed by dividing the panoramic image by a specific pitch angle. Specifically, there will be some repeated areas between two block images of adjacent pitch angles. The existence of these repeated areas is to prevent the lesions caused by the disease from being separated by adjacent block images, so that each block image only contains partial features. In other words, the repeated area can enable the lesions to be fully represented in a block image. Each block image is independently input into the neural network for analysis and processing, but these block images share the same acquisition position, and the GPS system on the drone 1 uniformly records the location information during acquisition.

[0065] like Figure 7 As shown, Figure 7 It is a schematic diagram of a block image. Figure 7 It is divided into a, b and c. A is a block image, b is the block image next to a, and c is a 360-degree panoramic image. It can be seen that there is a repeated area between the two block images. It should be noted that Figure 7 For illustration purposes only, it is rare that a panoramic image is divided into only two tile images.

[0066] In this embodiment, the block images may be directly acquired by the image acquisition device, or the panoramic image may be pre-processed to obtain the block images, which is not limited here.

[0067] The processing method using block images has the following advantages. First, the panoramic image may be too large and complex, so its processing speed is slow during subsequent identification. The block images can be analyzed in batches to increase the processing speed. Secondly, the block images can provide higher resolution and clearer local details, which is convenient for fine identification of diseases. In addition, by setting overlapping areas, the system can cross-verify the disease identification results, improve the reliability and accuracy of identification, and remove duplicate identification results through elimination methods, so that the content presented in the final disease distribution map is more accurate. Finally, the block images can realize parallel analysis, improve the overall processing efficiency, and shorten the data processing time.

[0068] The block images are then input into the disease recognition neural network for identification. Here, all the block images collected by a drone 1 in a mission area are taken as an example for explanation, as follows:

[0069] 1) Image Registration

[0070] Assuming that N block images are collected in a task area, the position information of these N block images is the same. Then the N block images collected are image registered. The goal of image registration is to uniformly map the angle, shape and size of each block image to the same coordinate system to achieve spatial alignment and consistency of multiple block images. Image registration is accomplished by matching the same-name points in the block images. The same-name points refer to the pixels that represent the same actual object or position in different images. For example, in two adjacent block images, a feature point on a rice leaf (such as a unique spot or a vein intersection point) appears in both block images at the same time. This feature point is the same-name point.

[0071] During the registration process, the algorithm identifies and matches some of the same points in each tile to achieve alignment and spatial consistency of the tiles. For example, methods such as the Scale Invariant Feature Transform (SIFT) algorithm can be used to identify the same points in different images. These same points are compared and matched using feature descriptors to ensure accurate alignment even if the tiles vary in angle or scale. In this way, the registration algorithm transforms each tile into the same coordinate system so that they can be analyzed in the same space.

[0072] 2) Neural network recognition

[0073] After image registration, the N block images will be preprocessed to obtain the preprocessed N block images. Then, the preprocessed N block images will be input into the disease recognition neural network to obtain the recognition results of all the disease points in the N block images.

[0074] The preprocessing includes normalization of the block images, feature recognition, feature importance recognition, and feature selectivity enhancement. The disease recognition neural network may include: YOLOv10 neural network. This will be described in detail later.

[0075] Image normalization: The purpose of image normalization is to standardize the brightness, contrast and other features of each block image to eliminate image differences caused by different lighting conditions. Through normalization, the data range of all block images is adjusted to the same scale, so that the subsequent disease recognition neural network can focus on the disease features during processing without being disturbed by external lighting or shooting conditions.

[0076] Image feature recognition: Using the preset perception layer, important features related to the disease in each block image can be extracted, such as the shape, texture and edge information of the lesion on the upper water surface. Feature recognition helps to highlight key areas in the image, making it easier for the subsequent neural network to locate and classify the degree of the disease.

[0077] Image feature importance identification: Feature importance identification is to analyze the extracted multiple image features to determine which features are most critical in disease detection. Through weighted calculation or gradient boosting and other methods, the system assigns importance weights to the color, edge shape, texture and other information of the disease features to identify the most representative and distinguishing features. This step ensures that the disease recognition model focuses more on the most effective features, thereby improving the efficiency and accuracy of detection.

[0078] Feature selective enhancement: Based on feature importance recognition, key features are selectively enhanced to make disease features more prominent in the image. Key features can be obtained through feature importance recognition, for example, the top M features in the feature importance recognition results are used as key features. Through image enhancement techniques such as histogram equalization and sharpening filtering, the selected disease features are made more prominent in contrast and clarity, which facilitates further detailed analysis by the neural network. Through selectively enhanced images, minor disease features, such as small-scale lesion expansion or slight color changes, can be more accurately detected and identified, thereby improving the sensitivity of the model in early disease recognition.

[0079] The N pre-processed block images are input into the disease recognition neural network. The neural network processes each image one by one and identifies the location, size and severity of all disease spots. The recognition result can be a grading result. For example, the disease grading standard in "NY / T 2720-2015 Technical Specifications for Identification of Rice Sheath Blight Resistance" is adopted to divide sheath blight into different types such as 0 (asymptomatic), 1 (mild), 3 (moderate) and 5 (severe) according to the size of the lesions and the spread of the disease. Of course, it can also be a two-level or other type of grading, which is not limited here.

[0080] 3) Deduplication of recognition results

[0081] In the above, the positions of all the diseased points are identified and marked. Then, it is determined whether the marks in different block images are repeated marks, and the repeated marks are merged based on the determination result to obtain a unique mark corresponding to a diseased position.

[0082] In some embodiments, in order to detect duplicate marks, the system will analyze these marks to identify which marks belong to the same map point. If they are the same map point, these marks are determined to be the same defect point. Next, the system will merge these duplicate marks, retaining only one unique mark to represent a defect point, while removing other redundant marks.

[0083] In other embodiments, a spatial clustering-based algorithm (e.g., DBSCAN) may be used to identify and merge close disease points. In addition, to ensure data integrity and prevent misjudgment, the system may also combine characteristic attributes of the disease points, such as color and texture features, to further verify the accuracy of deduplication.

[0084] Through this deduplication process, the system avoids counting the same disease point multiple times in overlapping areas, prevents the severity of the disease from being overestimated, and ensures the objectivity and reliability of disease statistics.

[0085] Step S30: Based on the location information, the recognition results are subjected to spatial distribution visualization processing to obtain a disease distribution map.

[0086] It has been explained in the above content that the acquired image corresponds to the corresponding position information. Similarly, the recognition result naturally corresponds to the same position information. In some embodiments, the position information is longitude and latitude information.

[0087] After deduplication, the recognition results are first integrated with the location information to associate the location information with its disease level. In this way, each identified disease point is accurately mapped to a specific location in the field. This process uses a mapping algorithm to match the mark of the recognition result with the location information to achieve accurate positioning of the disease in the production field.

[0088] Subsequently, the integrated data is visualized for spatial distribution to generate a disease distribution map. In this process, different colors can be used to encode different disease severity levels. For example, green is used to represent healthy areas, yellow is used to represent areas with minor diseases (level 1-3), and red is used to represent areas with severe diseases (level 4-5). Using visualization ensures that users can quickly identify high-risk areas when viewing the disease distribution map.

[0089] After the disease distribution map is completed, the identification results are output to the geographic information system (GIS) for visual display.

[0090] Disease distribution maps can not only be used as disease monitoring and assessment tools, but also can be used to develop precise disease prevention and control plans. For example, combined with disease distribution information, the system can recommend that areas with severe diseases give priority to the use of targeted pesticides, while reducing or avoiding the use of pesticides in healthy areas, thereby achieving the goal of precise drug use.

[0091] This step integrates location information and identification results, combines spatial distribution visualization technology and GIS display, and achieves accurate positioning of the disease and visualization of its severity.

[0092] On the other hand, a training method for a disease recognition neural network is provided, such as Figure 8 As shown, Figure 8 This is a flow chart of the training process of a disease recognition neural network. The details are as follows:

[0093] 1. Capture images. The technical solution used for capturing images is similar to that of step S10 and will not be described in detail here.

[0094] 2. Transmit and store the collected images. The collected images contain the location information of the captured images. The collected images can be copied and transmitted by wired or wireless means. In some embodiments, the collected images are input into a database.

[0095] 3. Manually annotate the collected images in the database, and classify the sheath blight into different types according to the lesion size and disease extension, such as grade 0 (asymptomatic), grade 1-3 (mild), and grade 4-5 (severe), according to the disease classification standards in "NY / T 2720-2015 Technical Specifications for Identification of Rice Sheath Blight Resistance". Use these grades as labels for the corresponding images to obtain image samples.

[0096] After preprocessing, the image samples are input into the disease recognition neural network, and the disease recognition neural network is trained by back propagation to finally obtain a trained disease recognition neural network. The disease recognition neural network is a YOLOv10 neural network.

[0097] In this process, preprocessing includes: normalization, quality control, data augmentation and cleaning, feature recognition, feature importance recognition and feature selectivity enhancement. The preprocessed image samples are then input into the disease recognition neural network. The disease recognition neural network is trained using activation functions, batch normalization functions, random dropout, full connection and other methods.

[0098] Quality control refers to the screening and filtering of the collected images during the image processing process to ensure that the data input to the neural network is clear, free of noise or other factors that affect recognition. High-quality image input helps improve the recognition accuracy of the model. Data augmentation is a technique for generating new images by transforming the original image (such as rotating, flipping, cropping, scaling, etc.). This method helps to increase the diversity of training data, thereby improving the generalization and robustness of the neural network model. Data cleaning refers to removing invalid, ambiguous, repeated or abnormal images from the data set to ensure the quality and consistency of the input data. The cleaned data set is more suitable for training and improving model performance. Batch normalization standardizes the input data of each layer in the neural network, thereby accelerating the training process and improving model stability. Batch normalization keeps the data evenly distributed during the training process by standardizing the input features in each small batch. Random dropout is a regularization technique used to prevent neural networks from overfitting. It forces the network to learn more feature combinations and improve the generalization ability of the model by randomly ignoring a part of neurons and their connections during the training process.

[0099] In the training process of the disease recognition neural network, the image samples that have been processed by normalization, quality control, data augmentation and cleaning, feature recognition, feature importance recognition, and feature selective enhancement are input into the disease recognition neural network, which can significantly improve the analysis accuracy and robustness of the model. By using activation functions, batch normalization functions, random inactivation, and full connection techniques in the training process, the neural network model is optimized, and the nonlinear expression ability, training stability, and generalization ability are improved, and finally the accurate recognition and classification of diseases are achieved, ensuring the reliability of the recognition results and the high efficiency in practical applications.

[0100] Furthermore, in the training process of the disease recognition neural network, the collected images may also be block images, which are labeled and called block image samples.

[0101] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0102] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes a computer program, and the computer program can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0103] Furthermore, the present invention also provides a control device. In an embodiment of a control device according to the present invention, the control device includes a processor and a storage device, the storage device can be configured to store a program for executing the method for identifying rice diseases using unmanned equipment in the above method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the method for identifying rice diseases using unmanned equipment in the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0104] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the method for identifying rice diseases using unmanned equipment in the above-mentioned method embodiment, and the program can be loaded and run by a processor to implement the above-mentioned method for identifying rice diseases using unmanned equipment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0105] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0106] Those skilled in the art will appreciate that the modules in the device can be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0107] It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications should also be regarded as the scope of protection of this application.

Claims

1. A method for identifying rice diseases using unmanned equipment, characterized in that: The unmanned equipment includes a drone, a connection mechanism, and an image collection device, wherein the image collection device and the drone are connected and distance-controlled via the connection mechanism, and the method includes: Obtaining a task instruction and sending the task instruction to a corresponding unmanned device, wherein the task instruction is used to control the unmanned device to be in a hovering state when the unmanned device reaches a task area above the rice, and to control the connection mechanism to extend so that the image collection device moves downward, passes through the leaf layer of the rice until it is close to the water surface, so as to obtain an image of the rice close to the water surface; The image is input into a trained disease recognition neural network to obtain a recognition result.

2. The method according to claim 1, characterized in that The method further comprises: Acquire location information corresponding to the recognition result, wherein the location information is used to represent a shooting location of the image; Based on the location information, the recognition results are processed for spatial distribution visualization to obtain a disease distribution map.

3. The method according to claim 1, characterized in that The step of obtaining an image of rice near the water surface comprises: Acquire N block images, wherein the N block images correspond to the same position information, the position information is used to represent the position of the drone when the block images are acquired, the N block images are used to represent images captured by the image acquisition device at different viewing angles, and there are repeated areas between two block images at adjacent viewing angles; The step of inputting the image into a trained disease recognition neural network to obtain a recognition result includes: Get N block images with the same position information; Acquire the same-name points of the N block images, and perform image registration on the N block images based on the image name points to obtain the N block images after registration, wherein the same-name points are used to represent pixel points representing the same actual position in different block images; Inputting the registered N block images into a disease recognition neural network to obtain recognition results of all disease points in the N block images, wherein the recognition results include marking the disease locations; The method further comprises: It is determined whether the marks in different block images are repeated marks, and the repeated marks are merged based on the determination result to obtain a unique mark corresponding to a disease location.

4. The method according to claim 3, characterized in that Determine whether the marks in different block images are repeated marks, including: It is determined whether the marks in different block images are points of the same name. If there are multiple marks that are points of the same name, the multiple marks are repeated marks.

5. A method for training a disease recognition neural network, characterized in that: include: Send different task instructions to a drone fleet, wherein the drone fleet includes a plurality of unmanned devices, any of the task instructions corresponds to one of the unmanned devices, the unmanned devices include a drone, a connecting mechanism, and an image collection device, wherein the image collection device and the drone are connected and distance-controlled via the connecting mechanism, wherein the task instruction is used to control the drone to be in a hovering state when the drone reaches a task area above the rice, and to control the connecting mechanism to extend so that the image collection device moves downward, passes through a leaf layer of the rice until it is close to the water surface, so as to obtain an image of the rice close to the water surface; After labeling the image, an image sample is obtained; The image samples are preprocessed, and the preprocessed image samples are input into a disease recognition neural network to obtain a recognition result, and the disease recognition neural network is trained using a back propagation algorithm based on the recognition result and the label of the image sample.

6. The method according to claim 5, characterized in that The preprocessing includes: normalization of image samples, quality control, data augmentation and cleaning, feature recognition, feature importance recognition and feature selectivity enhancement; the methods used in training the disease recognition neural network include: activation function, batch normalization function, random inactivation and full connection.

7. The method according to claim 5, characterized in that The step of obtaining an image of rice near the water surface comprises: Acquire N block images, wherein the N block images correspond to the same position information, the position information is used to represent the position of the drone when the block images are acquired, the N block images are used to represent images captured by the image acquisition device at different viewing angles, and there are repeated areas between two block images at adjacent viewing angles; The pre-processed image samples are input into the disease recognition neural network to obtain recognition results, including: Obtain N block image samples with the same position information after preprocessing; Acquire the same-name points of the N block image samples, and perform image registration on the N block image samples based on the same-name points to obtain the N block image samples after registration, wherein the same-name points are used to represent pixel points representing the same actual position in different block image samples; Inputting the registered N block image samples into the disease recognition neural network to obtain the recognition results of all the disease points in the N block image samples, wherein the recognition results include marking the disease locations; The method further comprises: It is determined whether the marks in different block image samples are repeated marks, and the repeated marks are merged based on the determination result to obtain a unique mark corresponding to a disease location.

8. An unmanned device, characterized in that: include: A drone and an image collection device, wherein the image collection device and the drone are connected via a connection mechanism, and the length of the connection mechanism can be changed to adjust the distance between the drone and the image collection device; The unmanned equipment also includes a floating device and a weight, wherein the floating device is used to support the image collecting device and make the image collecting device float on the water surface, the weight is located at the lower part of the floating device, and the upper limit of the buoyancy provided by the floating device is greater than the sum of the gravity of the image collecting device and the weight.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method of any one of claims 1 to 4 or claims 5 to 7 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method of any one of claims 1 to 4 or claims 5 to 7 through the computer program.

Images