Portable push-pull type rice pest image acquisition and recognition system

Abstract

The invention discloses a portable push-pull type rice pest image acquisition and recognition system, which comprises a portable box body arranged in a pest situation detection device, and the portable box body is in sliding connection with the pest situation detection device; a turntable is arranged at the bottom of the portable box body, is rotationally connected with the bottom of the portable box body and is used for bearing inactivated pests, an image acquisition device is arranged at the top of the turntable correspondingly and is electrically connected with a controller, and the controller is used for receiving images acquired by the image acquisition device to identify the rice pests; and the pests on the turntable are cleaned to the bottom of the portable box body after being collected once. The portable box body can be flexibly detached from the pest situation detection device, the internal rotating disc can also regularly clean the pests, and when the pests in the portable box body are accumulated to a certain degree, the portable box body can be removed to clean the pests, so that the influence of accumulation of a large number of pests is reduced, and the accuracy of subsequent pest recognition is improved.

Description

Technical Field

[0001] The present application relates to the technical field of rice pest identification, and in particular to a portable push-pull rice pest image acquisition and identification system. Background Art

[0002] In recent years, rice production has been facing the situation of outbreaks, recurrences and frequent occurrences of pest groups, which has become an important factor restricting the safe production of rice. The threat of pests such as brown planthoppers, gray planthoppers, white-backed planthoppers and rice leaf rollers poses a serious threat to the rice industry and brings great pressure to agricultural production. Therefore, timely detection of the occurrence dynamics and degree of damage of these pests and the formulation of scientific and effective prevention and control strategies are key measures to ensure the safety of rice production and promote the sustainable development of the industry.

[0003] The traditional technology of relying on manual pest trapping and identification is easily affected by the observer's personal experience and subjective consciousness, resulting in inconsistency and low accuracy of the detection results. Chinese patent CN118452170A discloses a live rice pest targeted trapping system and identification method. During the trapping process, the rice pests are detected by an infrared detection device and fed back to the controller, which triggers the image acquisition device to shoot and identify the target rice pests. Through targeted identification and accurate positioning of rice pests, the live pests are identified and counted in real time using a deep learning algorithm, and the pests are tracked and photographed by an infrared detection device and an industrial camera, which improves the trapping efficiency and reduces the impact of non-target insects.

[0004] The above patent can achieve more accurate trapping and identification of rice pests compared to traditional technologies, but the pest trapping and image acquisition are fixed. As the monitoring time increases, more and more pests will accumulate, which will affect the accuracy of subsequent pest identification. Summary of the invention

[0005] In order to solve the above technical problems, this application proposes the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a portable push-pull rice pest image acquisition and identification system, comprising: a portable case arranged in an insect detection device, the portable case being slidably connected to the insect detection device; a turntable is arranged at the bottom of the portable case, the turntable is rotatably connected to the bottom of the portable case, the turntable is used to carry inactivated pests, an image acquisition device is arranged corresponding to the top of the turntable, the image acquisition device is electrically connected to a controller, the controller is used to receive images acquired by the image acquisition device for rice pest identification, and after one acquisition, the pests on the turntable are cleaned to the bottom of the portable case.

[0007] In a possible implementation, a sliding track is provided at the bottom of the insect detection device case, and a pulley is provided at the bottom of the portable case, and the pulley is slidably connected to the sliding track. The portable case is equipped with two sets of pulleys, and the sliding track is provided accordingly. The lower end of the portable case is connected to the sliding track through the groove of the pulley, so that it can slide smoothly. In order to prevent the portable case from accidentally sliding when not in operation, removable fixing holes are provided at both ends of the outer side of the sliding track so that the portable case can be fixed when needed.

[0008] In a possible implementation, a ring-shaped LED light strip is provided on the top of the portable box, and the inner wall of the portable box and the surface of the turntable are coated with a pure black paint with high light absorption rate. The ring-shaped LED light strip is installed on the top of the portable box to illuminate the turntable, and the ring-shaped design reduces the interference of different light strip position settings on the image acquisition device. The pure black paint with high light absorption rate is applied to reduce the interference of the reflected light from the inner wall of the box on the image acquisition device. The ring-shaped light strip and the pure black paint synergistically improve the clarity and feature recognition of the collected rice pest images.

[0009] In a possible implementation, the surface of the turntable is coated with a pure black paint with high light absorption rate, and then white scale lines are drawn. The scale lines are drawn on the turntable with white paint to further improve the clarity and feature recognition of the collected rice pest images.

[0010] In a possible implementation, a first motor is disposed at the bottom of the turntable, and a rotation output end of the first motor is fixedly connected to the turntable; a disc brush is disposed on the turntable, and one end of the disc brush is fixedly connected to a rotation output shaft of the second motor. After the image acquisition device acquires the image of the inactivated rice pests, it transmits a signal to the controller, and the controller controls the first motor and the second motor to work. When the turntable rotates, the disc brush also starts to clean the inactivated rice pests on the turntable and remove them.

[0011] In a possible implementation, a transmission channel is provided on one side of the portable box, and the transmission channel is connected to the insect detection device.

[0012] In a possible implementation, the controller receives the image captured by the image acquisition device to identify rice pests, including: a YOLO-v8-based deep learning system is provided in the controller, and the deep learning system identifies and counts the target rice pests according to the image.

[0013] In a possible implementation, the deep learning system identifies and counts target rice pests according to the image, including:

[0014] Apply the CBAM attention module to the YOLO-v8 model, wherein the CBAM attention module includes a channel attention module CAM and a spatial attention module SAM;

[0015] In CAM, global average pooling and global maximum pooling operations are performed on the input rice pest feature map to obtain a vector of two channel dimensions;

[0016] Then the two vectors are combined through the sigmoid function to generate the final channel attention map;

[0017] In SAM, the feature map is operated in spatial dimension to capture key information at different spatial locations;

[0018] After the pests are identified, the target image is processed for detail features. The features in the image are extracted through the image mask. The pre-made pest gender feature interest region mask is multiplied with the image to be processed to obtain the interest region image. The image value in the interest region remains unchanged, while the image values ​​outside the region are all 0.

[0019] Use masks to shield certain areas on the image so that they do not participate in the processing or the calculation of processing parameters, or only process or count the shielded areas to select the sex of the pests in a targeted manner.

[0020] In a second aspect, an embodiment of the present application provides a rice pest identification method based on the portable push-pull rice pest image acquisition and identification system described in any possible implementation of the first aspect, comprising:

[0021] When the inactivated rice pests enter the portable box and slide onto the rotating disk, the image acquisition device takes a picture of the rice pests;

[0022] After the image data captured by the image acquisition device is transmitted to the controller, the built-in deep learning model performs analysis and processing to complete the pest identification statistics;

[0023] When the preset time is reached, the controller controls the disc brush to clean the inactivated rice pests on the turntable.

[0024] In an embodiment of the present application, the portable box can be flexibly removed from the insect detection device, and the internal turntable can also regularly clean the pests. When the pests in the portable box accumulate to a certain extent, the portable box can be removed to clean the pests, thereby reducing the impact of a large amount of pest accumulation and improving the accuracy of subsequent pest identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of a portable push-pull rice pest image acquisition and identification system provided in an embodiment of the present application;

[0026] Figure 2 A physical picture of the insect detection device provided in the embodiment of the present application;

[0027] Figure 3 A schematic diagram of a pulley track provided in an embodiment of the present application;

[0028] Figure 4 A schematic diagram of a turntable provided in an embodiment of the present application;

[0029] Figure 5 A schematic diagram of a process for identifying rice pests provided in an embodiment of the present application;

[0030] Figure 1-Figure 5 In the symbol, it is represented as:

[0031] 1-portable box, 2-transmission channel, 3-turntable, 4-image acquisition device, 5-controller, 6-sliding track, 7-pulley, 8-ring LED light strip, 9-first motor, 10-disc brush, 11-second motor. DETAILED DESCRIPTION

[0032] The present solution is described below in conjunction with the accompanying drawings and specific implementation methods.

[0033] See also Figure 1 The portable push-pull rice pest image acquisition and identification system provided in the embodiment of the present application includes: a portable case 1 arranged in an insect detection device, and the portable case 1 is slidably connected to the insect detection device. A transmission channel 2 is arranged on one side of the portable case 1, and the transmission channel 2 is connected to the insect detection device. A turntable 3 is arranged at the bottom of the portable case 1, and the turntable 3 is rotatably connected to the bottom of the portable case 1. The turntable 3 is used to carry inactivated pests. An image acquisition device 4 is arranged corresponding to the top of the turntable 3. The image acquisition device 4 is electrically connected to a controller 5. The controller 5 is used to receive the image acquired by the image acquisition device 4 for rice pest identification. After one acquisition, the pests on the turntable 3 are cleaned to the bottom of the portable case 1.

[0034] join Figure 2 This is a physical picture of the insect detection device. Since the traditional photo box is fixed inside the insect detection device, it is very inconvenient to push and pull, and it is impossible to observe the specific usage of the internal device. Therefore, a slide rail device is added to make the rice pest photo box in this example different from the traditional rice pest photo box. It is more convenient to observe and replace equipment, and it can be used normally even in harsh environments.

[0035] Specifically, see Figure 3In this embodiment, the bottom of the insect detection device case is provided with a sliding track 6, and the bottom of the portable case 1 is provided with a pulley 7, and the pulley 7 is slidably connected with the sliding track 6. The portable case 1 is equipped with two sets of pulleys 7, and the sliding track 6 is correspondingly provided. The lower end of the portable case 1 is connected to the sliding track 6 through the groove of the pulley 7, so that it can slide smoothly. In order to prevent the portable case 1 from accidentally sliding in a non-working state, detachable fixing holes are provided at both ends of the outer side of the sliding track 6 so that the portable case 1 can be fixed when necessary.

[0036] The image acquisition device 4 in this embodiment selects a deep industrial camera with high-resolution image acquisition capability for shooting, which can achieve high-precision visual inspection and measurement in fine details, and can take pictures of the inactivated rice pests that fall on the turntable 3. At the same time, the camera has fast image acquisition features, including short exposure time and high-speed image transmission, which improves efficiency and reduces errors. It can take detailed pictures of inactivated rice pests, especially the sexual characteristics of inactivated pests. The use of high-sensitivity and low-noise sensor technology enables the industrial camera to obtain clear images under low light conditions. It can be used normally even in harsh rice field environments.

[0037] In this embodiment, the outer frame of the portable box 1 is in the shape of a cube, and the length, height and width of the outer frame of the portable box 1 are the same. A layer of material with flexibility, resistance to environmental stress cracking and good weather resistance is fixedly installed on the inner wall of the portable box 1. The Shore hardness of the material is set to 70 degrees, which is used to provide buffering and protection, alleviate the fluctuation caused by the sliding of the box when it is taken out, and protect the controller 5 and the industrial camera from damage.

[0038] A ring-shaped LED light strip 8 is arranged on the top of the portable box 1, and the inner wall of the portable box 1 and the surface of the turntable 3 are coated with a pure black paint with high light absorption rate. The ring-shaped LED light strip 8 is installed on the top of the portable box 1 to illuminate the turntable 3, and the ring-shaped design reduces the interference of different light strip position settings on the image acquisition device 4. The pure black paint with high light absorption rate is applied to reduce the interference of the reflected light on the inner wall of the box on the image acquisition device 4. The surrounding back panels can effectively enhance the outline and features of the photographed object, making it more prominent, and effectively reduce the interference of stray light in the image. The ring-shaped light strip and the pure black paint synergistically improve the clarity and feature recognition of the collected rice pest images, making full preparations for the subsequent identification of pests.

[0039] Furthermore, the surface of the rotating disk 3 is coated with a pure black paint with high light absorption rate, and then white scale lines are drawn. The scale lines are drawn on the rotating disk 3 with white paint to further improve the clarity and feature recognition of the collected rice pest images.

[0040] See also Figure 1 and Figure 4 The first motor 9 is arranged at the bottom of the rotating disk 3, and the rotation output end of the first motor 9 is fixedly connected to the rotating disk 3; the rotating disk 3 is provided with a disc brush 10, and one end of the disc brush 10 is fixedly connected to the rotation output shaft of the second motor 11. After the image acquisition device 4 acquires the inactivated rice pests picture, it transmits the signal to the controller 5, and the controller 5 controls the first motor 9 and the second motor 11 to work. When the rotating disk 3 rotates, the disc brush 10 also starts to clean the inactivated rice pests on the rotating disk 3 and remove them.

[0041] Due to the vast area of ​​farmland, in order to avoid the problem that rice pests cannot be continuously photographed and observed after accumulating to a certain number in the device, this example leaves space under the turntable 3 to treat and inactivate rice pests. The disc brush 10 and the turntable 3 rotate to slide the pests to the bottom of the cabin for accumulation. Every 12 hours, the controller 5 promptly notifies the staff to carry out corresponding treatment, thereby improving the efficiency of pest trapping and treatment and reducing labor and time costs.

[0042] In this embodiment, YOLO-v8 deep learning is used to identify target pests, combined with transfer learning and data enhancement technology to improve the accuracy and generalization ability of target pests. Prior to this, in order to improve the performance of convolutional neural networks, a large separable kernel CBAM attention module was first introduced into the target detection algorithm. This module contains two important components: channel attention module (CAM) and spatial attention module (SAM). First, in CAM, the input rice pest feature map is subjected to global average pooling and global maximum pooling operations to obtain vectors of two channel dimensions. Then, the two vectors are combined by the sigmoid function to generate the final channel attention map. In SAM, the feature map is operated in spatial dimensions to capture key information at different spatial positions. CBAM applies CAM and SAM to the input feature map in turn to achieve attention adjustment in two dimensions of channel and space, thereby emphasizing feature channels with high discrimination, suppressing redundant and useless features, and accurately extracting key features of rice pest images, such as color, shape, and texture. By dynamically adjusting the weights of different channels in the feature map, CBAM can adapt to various complex scenes more effectively. In addition, CBAM also introduces adaptive average pooling and adaptive maximum pooling to calculate channel weights, so that the network can fully consider the average and maximum values ​​of the input feature map. In this embodiment, the CBAM attention module is applied to the YOLO-v8 model. The results show that it can significantly improve the MAP index of the model on the data set, bringing better performance to the target detection task.

[0043] The comprehensive application of the above data enhancement techniques can generate a large number of diverse and rich training samples without changing the semantic information of the original image, thereby improving the performance and robustness of the model in target detection tasks. By combining the CBAM attention module with transfer learning and data enhancement techniques, the performance of convolutional neural networks in target detection tasks can be effectively improved. This method can not only achieve more accurate target detection on limited pest datasets, but also adapt to the needs of various complex scenarios, providing strong support for research and application in the field of rice pest detection.

[0044] After YOLO-v8 finishes identifying pests, the characteristic images of the rice leaf folder and the striped stem borer under study are extracted, and then the target image is processed for detail features, and the features in the image are extracted through the image mask. The pre-made rice leaf folder and striped stem borer gender characteristic interest area mask is multiplied with the image to be processed to obtain the interest area image, and the image value in the interest area remains unchanged, while the image value outside the area is 0. Use the mask to shield certain areas on the image so that they do not participate in the processing or the calculation of the processing parameters, or only process or count the shielded area, and select the sex of the pests in a targeted manner.

[0045] See also Figure 5 , is a rice pest identification method using the portable push-pull rice pest image acquisition and identification system of the above embodiment, comprising:

[0046] S101, when the inactivated rice pests enter the portable box 1 and slide onto the rotating disk 3, the image acquisition device 4 takes a photo of the rice pests.

[0047] S102, after the image data captured by the image acquisition device 4 is transmitted to the controller 5, the built-in deep learning model performs analysis and processing to complete the pest identification statistics.

[0048] S103, when the preset time is reached, the controller 5 controls the disc brush 10 to clean the inactivated rice pests on the rotating disc 3.

[0049] In the embodiments of the present application, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0050] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A portable push-pull rice pest image acquisition and identification system, characterized in that: include: A portable box is arranged in an insect detection device, and the portable box is slidably connected to the insect detection device; a turntable is arranged at the bottom of the portable box, and the turntable is rotatably connected to the bottom of the portable box, and the turntable is used to carry inactivated pests. An image acquisition device is arranged at the top of the turntable, and the image acquisition device is electrically connected to a controller. The controller is used to receive images acquired by the image acquisition device for rice pest identification. After one acquisition, the pests on the turntable are cleaned to the bottom of the portable box.

2. The portable push-pull rice pest image acquisition and identification system according to claim 1, characterized in that: A sliding track is arranged at the bottom of the insect detection device box, and a pulley is arranged at the bottom of the portable box, and the pulley is slidably connected with the sliding track.

3. The portable push-pull rice pest image acquisition and identification system according to claim 1, characterized in that: A ring-shaped LED light strip is arranged on the top of the portable box, and the inner wall of the portable box and the surface of the turntable are coated with pure black paint with high light absorption rate.

4. The portable push-pull rice pest image acquisition and identification system according to claim 3, characterized in that: The surface of the turntable is coated with a pure black paint with high light absorption rate, and then white scale lines are engraved.

5. The portable push-pull rice pest image acquisition and identification system according to claim 1, 3 or 4, characterized in that: A first motor is arranged at the bottom of the turntable, and a rotation output end of the first motor is fixedly connected to the turntable; a disk brush is arranged on the turntable, and one end of the disk brush is fixedly connected to a rotation output shaft of the second motor.

6. The portable push-pull rice pest image acquisition and identification system according to claim 1, characterized in that: A transmission channel is provided on one side of the portable box, and the transmission channel is connected to the insect detection device.

7. The portable push-pull rice pest image acquisition and identification system according to claim 1, characterized in that: The controller receives the image collected by the image collection device to identify rice pests, including: a YOLO-v8-based deep learning system is set in the controller, and the deep learning system identifies and counts the target rice pests according to the image.

8. The portable push-pull rice pest image acquisition and identification system according to claim 7, characterized in that: The deep learning system identifies and counts target rice pests according to the image, including: Apply the CBAM attention module to the YOLO-v8 model, wherein the CBAM attention module includes a channel attention module CAM and a spatial attention module SAM; In CAM, global average pooling and global maximum pooling operations are performed on the input rice pest feature map to obtain a vector of two channel dimensions; Then the two vectors are combined through the sigmoid function to generate the final channel attention map; In SAM, the feature map is operated in spatial dimension to capture key information at different spatial locations; After the pests are identified, the target image is processed for detail features. The features in the image are extracted through the image mask. The pre-made pest gender feature interest region mask is multiplied with the image to be processed to obtain the interest region image. The image value in the interest region remains unchanged, while the image values ​​outside the region are all 0. Use masks to shield certain areas on the image so that they do not participate in the processing or the calculation of processing parameters, or only process or count the shielded areas to select the sex of the pests in a targeted manner.

9. A rice pest identification method based on the portable push-pull rice pest image acquisition and identification system according to any one of claims 1 to 8, comprising: When the inactivated rice pests enter the portable box and slide onto the rotating disk, the image acquisition device takes a picture of the rice pests; After the image data captured by the image acquisition device is transmitted to the controller, the built-in deep learning model performs analysis and processing to complete the pest identification statistics; When the preset time is reached, the controller controls the disc brush to clean the inactivated rice pests on the turntable.

Images