PANet algorithm-combined greenhouse crop growth environment accurate adjustment method
Through the combination of PANet algorithm and PID controller, the precise temperature regulation of a single area in the greenhouse is achieved by using temperature sensors and infrared detectors, which solves the problem of inaccurate temperature regulation in the existing technology, ensures that the crops grow in an ideal environment, and achieves the Intelligent temperature control and resource conservation.
Patent Information
- Application Number
- CN202411141669.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art is difficult to achieve single-region temperature regulation in greenhouses, resulting in the inability to accurately reach the ideal equilibrium value, affecting crop growth.
The PANet algorithm is used to combine temperature sensors and infrared temperature detectors to achieve accurate temperature regulation of a single target area in the greenhouse through the environmental dynamic differential equation and PID controller. The temperature sensor group is evenly distributed and uploaded to the Internet of Things cloud platform. The infrared image is divided into segments and area information is obtained by combining the PANet algorithm, and feedback adjustment is performed with the PID controller to ensure that the temperature reaches the ideal growth range.
It realizes accurate temperature regulation in a single area in the greenhouse, ensures that the crops grow in an ideal growth environment, reduces labor costs, and realizes intelligent unmanned temperature control, saves resources and improves crop yield and quality.
Smart Images

Figure CN119002602A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of temperature control, and in particular to a temperature control method for an intelligent greenhouse. Background Art
[0002] In the current process of scientific and technological development, the application of smart greenhouses is becoming more and more extensive, and more and more greenhouse plantings have achieved intelligent control. However, there are still many problems in the existing technology. It is impossible to accurately find the pathological area of the greenhouse to adjust the temperature to the target temperature balance value and make the greenhouse temperature reach the ideal temperature. It is also impossible to achieve single-region temperature regulation. Nowadays, the technology in the field of computer vision is developing more and more comprehensively and more and more intelligently. The cross-domain combination of computer vision and greenhouse temperature control can solve the problem and realize intelligent temperature regulation and control. Summary of the invention
[0003] The technical problem to be solved by the present invention is how to achieve single-zone temperature regulation, so as to adjust the overall temperature of each position in the greenhouse to an ideal temperature balance value, so that crops can grow in a more ideal temperature environment.
[0004] The present invention solves the above technical problems through the following technical means: a method for accurately adjusting the growth environment of greenhouse crops combined with the PANet algorithm, comprising the following steps:
[0005] S1. Arrange at least one set of temperature sensors at different locations in the greenhouse to sense the greenhouse temperature, and use DTU to upload the data and compare it with the ideal temperature range for crop growth;
[0006] S2, using an infrared temperature detector to extract infrared images of greenhouse crops, segmenting the extracted infrared images of greenhouse crops through the PANet algorithm, dividing the greenhouse into several areas, and finally obtaining mask information including position and edge information;
[0007] S3. Calculate the change in greenhouse ambient temperature through the environmental dynamic differential equation, and combine it with the feedback control method of the PID controller to regulate the greenhouse area that needs to be regulated, and accurately adjust the temperature of a single target area. After the adjustment is completed, use regional image segmentation and temperature sensor to sense again to obtain data feedback to observe whether secondary regulation is needed to achieve stable control.
[0008] Furthermore, the S1 includes:
[0009] S11, firstly, evenly distribute the temperature sensor groups at the same interval positions in the greenhouse;
[0010] S12, multiple sensors collect temperature data from different planting sites simultaneously but the temperature data transmission intervals are different;
[0011] S13, all sensor groups are connected to a DTU respectively to upload the collected temperature data to the IoT cloud platform;
[0012] S14. Through the processor, the temperature data obtained by the IoT cloud platform is compared with the ideal growth temperature range of the crop to check whether the environment in which the crop is located is conducive to its growth. The sensors in the same sensor group are numbered, and the upload delay of the front and rear sensors is set, and the upload frequency of the front and rear sensors is the same.
[0013] Further, the S2 includes:
[0014] S21, feature extraction, after obtaining a pre-processed image, PANet will make ResNet perform preliminary multi-scale feature extraction;
[0015] S22, feature fusion, then use top-down and bottom-up feature fusion methods to fuse the multi-scale feature maps, and then use adaptive feature pooling to downsample the obtained region proposals for input into the detection head;
[0016] S23, obtaining the region of interest. In the detection head, a series of convolution operations are performed on the input feature map to obtain the target category information and mask information. This image uses different colors to represent different temperatures. The integration together is a pseudo-color map of the surface temperature distribution of the detected crop and its environment. If the crop environment temperature is in good condition, the detected infrared image is relatively uniform. Otherwise, the infrared image will present an uneven color picture.
[0017] S24, obtaining detection head results.
[0018] Furthermore, in the S21 feature extraction, after obtaining a pre-processed image, PANet causes ResNet to perform preliminary multi-scale feature extraction and obtain multiple multi-scale feature maps.
[0019] Furthermore, in the S21 feature fusion, the feature map extracted by ResNet is used to construct a feature pyramid, and an additional bottom-up feature enhancement path is constructed for the feature pyramid. In this path, the feature map is fused from bottom to top, from large size to small size, so that the high-level feature map obtains detailed information on the original basis. PANet uses a shallower N i And deeper P i+1 The fusion method gets its next layer N i+1 , it first passes a 3*3 convolution with a stride of 2 on N i Downsample, and then add P i+1The feature map after downsampling is then fused, and then a 3*3 convolution is used to fuse the features to increase the representation ability of the fused features, and finally the ReLU activation function is used to nonlinearize the features;
[0020] After obtaining the multi-level feature map N, RPN is used to classify the foreground and background in the feature map. RPN will generate a series of anchor frames in each level of the feature map, and score and fine-tune the size of each anchor frame through regression and classification. Finally, a set of anchor frame coordinates will be output for each feature map for subsequent operations. However, the scales of the feature maps at each level are different, and the global and feature information they contain are different, which makes each feature map contribute differently to the target segmentation. Based on the above, adaptive feature pooling is used for feature fusion.
[0021] Furthermore, S23, obtaining the region of interest is specifically as follows:
[0022] The feature map Φ obtained after each ROI fusion will be input into the detection head as the input feature map to output the result. Regarding the generation of the target mask, the feature map Φ needs to be convolved three times to obtain Φ'. Then, it will enter two branches, one of which uses FCN to generate the corresponding segmentation mask M for each category; the other branch inputs the feature map Φ' into two fully connected layers, and outputs a feature vector V containing the foreground and background binary classification information of each point in the feature map. Then, the feature vector V and the segmentation mask M are resized to make them the same size as Φ'. The resize correction operation of M is performed using deconvolution upsampling, and the feature map with more pixel-level details of the foreground and background is fused with the mask obtained by FCN to obtain a high-precision segmentation mask of the final target. After segmentation, the object to be tested is extracted. The infrared image carries the object temperature information, and the temperature detection information corresponding to the segmented target can be extracted. At the same time, the coordinate position of the segmented crop can be obtained for subsequent precise fixed-point control.
[0023] Furthermore, in S3, in calculating the change in the temperature of the greenhouse environment by using the dynamic differential equation of the environment, the dynamic differential equation of the greenhouse environment is:
[0024]
[0025] Where X m is the m-dimensional indoor environment variable; U n is the n-dimensional greenhouse control equipment state variable; V o is o-dimensional crop physiological state variable; D p is the p-dimensional outdoor environment disturbance variable; C q is the q-dimensional greenhouse structure parameter; t is time, s; f is a nonlinear function based on energy and material balance;
[0026] Furthermore, in S3, the feedback control method of the PID controller is combined to control the greenhouse area that needs to be regulated, and the temperature of a single target area is accurately adjusted, including:
[0027] After the change in the ambient temperature in the greenhouse is fed back to the PID control module through the processor, the environmental control method is gradually transformed into a dynamic control method based on the set value and real-time feedback from environmental monitoring. The system error is feedback controlled through proportional, integral and combined environmental dynamic differential equation operations:
[0028] The output u(t) of the PID control for regional regulation is:
[0029]
[0030] Where e(t) is the difference between the actual measured value and the set value, K p , T1, T D They are proportional gain, integral time constant and differential time constant respectively. The PID controller adjusts the three parameters K p , T1, T D That is, it can improve the dynamic and static characteristics of the system, and then use smart devices to perform small-scale temperature control based on the target area points obtained from the previous infrared images to make the target area reach the ideal growth temperature range.
[0031] Furthermore, the environmental control method is specifically as follows:
[0032] Combined with the dynamic differential equation of the greenhouse environment, PID control can then ensure the stable control of the greenhouse environment through feedback adjustment. The average standard error of the adaptive control strategy is reduced. The system selects the model category according to the input and output data, and uses system identification to adaptively adjust the unknown parameters in the model to ensure that the actual temperature gradually approaches the set value.
[0033] Further, it is applied to a regulation system including the following modules:
[0034] The ambient temperature sensing module includes at least one set of temperature sensors, which are arranged at different locations in the greenhouse to sense the greenhouse temperature at different locations and at different time points.
[0035] The infrared temperature tester is set up in the greenhouse to obtain infrared images of crops.
[0036] Data transmission module, using DTU to transmit temperature data and infrared images to the cloud platform, and connect relays for remote control;
[0037] The control and adjustment module uses the PANet algorithm to segment the extracted infrared image of greenhouse crops, divides the greenhouse into several areas, and finally obtains mask information including position and edge information; calculates the change in greenhouse ambient temperature through the environmental dynamic differential equation, and combines the feedback control method of the PID controller to regulate the greenhouse area that needs to be regulated, and accurately adjusts the temperature of a single target area. After the adjustment is completed, the regional image segmentation and temperature sensor are used to perceive again to obtain data feedback to observe whether secondary regulation is needed to achieve stable control.
[0038] It can be seen from the above technical scheme that the intelligent greenhouse temperature control method based on computer vision crop growth of the present invention, through the dual temperature acquisition method of the sensor and the infrared temperature detector, effectively and timely obtains the temperature environment of the greenhouse crop growth at a certain moment, and then obtains the target area position that does not meet the ideal growth environment interval through the approximate positioning of the infrared temperature detector, and realizes precise control of the temperature of the target crop environment by combining the environmental dynamic differential equation and PID control, so that the crops can reach the ideal growth environment as much as possible. Sudden changes in the environment around the greenhouse can also be easily solved by intelligent control, which reduces the labor cost, realizes unmanned and intelligent intelligent adjustment, and enables crops to achieve mass production and quality. The PANet algorithm can accurately obtain the precise position of the crop that needs to adjust the environment to achieve single pathological area adjustment, and the adjustment within the unnecessary range affects the growth of other crops in the normal growth environment. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 This is a flow chart of the method for accurately adjusting the growth environment of greenhouse crops in combination with the PANet algorithm of the present invention. DETAILED DESCRIPTION
[0040] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. The illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention and are not drawn according to the number, shape and size of components in actual implementation. The type, quantity and proportion of each component in actual implementation can be changed arbitrarily, and the component layout type may also be more complicated.
[0042] A method for accurately adjusting the growth environment of greenhouse crops in combination with the PANet algorithm, the method comprising the following steps:
[0043] S1. Arrange at least one set of temperature sensors at different locations in the greenhouse to sense the greenhouse temperature, and use DTU (Data Transfer Unit, wireless terminal device) to upload the data and compare it with the ideal temperature range for crop growth;
[0044] S2. Use an infrared temperature detector to extract infrared images of greenhouse crops, segment the extracted infrared images of greenhouse crops through the PANet (Path Aggregation Network) algorithm, divide the greenhouse into several areas, and finally obtain mask information including position and edge information;
[0045] S3. Calculate the change in greenhouse ambient temperature through the environmental dynamic differential equation, and combine it with the feedback control method of the PID controller to regulate the greenhouse area that needs to be regulated, and accurately adjust the temperature of a single target area. After the adjustment is completed, use regional image segmentation and temperature sensor to sense again to obtain data feedback to observe whether secondary regulation is needed to achieve stable control.
[0046] As a specific implementation of the present invention, S1 includes:
[0047] S11, first, evenly distribute the temperature sensor groups at the same interval positions in the greenhouse, for example, install one sensor at each of the two positions in front and behind the greenhouse planting area, which constitutes a sensor group;
[0048] S12, multiple sensors collect temperature data from different planting sites simultaneously but at different temperature data transmission intervals, such as the above two sensors collect the temperatures of adjacent planting sites respectively;
[0049] S13, all sensor groups are connected to a DTU respectively to upload the collected temperature data to the IoT cloud platform;
[0050] S14. Compare the temperature data obtained by the IoT cloud platform with the ideal growth temperature range of the crop through the processor to check whether the environment in which the crop is located is conducive to its growth. The sensors in the same sensor group are numbered. For example, the front sensor and the rear sensor of the first group are 1 and 2 respectively. A and 1 B ,The upload delay of the front and rear sensors is set to 30s with the same frequency, that is, each time the perceived temperature data is sent to the cloud platform via DTU at the same frequency but in different batches, thus achieving the role of monitoring and feedback.
[0051] As a specific embodiment of the present invention, in S2, segmenting the extracted greenhouse crop infrared image by the PANet algorithm, and finally obtaining mask information including position and edge information mainly comprises the following steps:
[0052] S21, feature extraction. After obtaining a pre-processed image, PANet causes ResNet to perform preliminary multi-scale feature extraction and obtain multiple multi-scale feature maps. As an embodiment of the present invention, four feature maps {R2, R3, R4, R5} of levels 2 to 5 in the ResNet output feature map are selected for a series of subsequent operations.
[0053] S22, feature fusion, uses top-down and bottom-up feature fusion methods to fuse multi-scale feature maps, and then uses adaptive feature pooling to downsample the obtained region proposals for input into the detection head;
[0054] As an embodiment of the present invention, a feature pyramid is constructed using feature maps {R2, R3, R4, R5} extracted by ResNet. There are two paths for constructing the feature pyramid, namely, horizontal feature connection and top-down feature fusion. Due to the limitation of the size of high-level feature maps, a large amount of detail information is usually lost in the process of feature extraction. For this reason, an additional bottom-up feature enhancement path is constructed for the feature pyramid. In this path, the feature maps are fused from bottom to top, from large size to small size, so that the high-level feature maps can obtain detail information on the original basis, which can improve the model's recognition and segmentation performance of local details. PANet uses a shallower N i And deeper P i+1 The fusion method gets its next layer N i+1 , it first passes a 3*3 convolution with a stride of 2 on N i Downsample, and then add P i+1 The feature map is then fused with the downsampled feature map, and then a 3*3 convolution is used to fuse the features to increase the representation ability of the fused features. Finally, the ReLU activation function is used to nonlinearize the features.
[0055] After obtaining the multi-level feature map N, RPN (Region Proposal Network) is used to classify the foreground and background in the feature map. RPN will generate a series of anchor boxes in each level of feature map, and score and fine-tune the size of each anchor box through regression and classification, and finally output a set of anchor box coordinates for each feature map for subsequent operations. However, the scales of feature maps at each level are different, and the global and feature information they contain are different, which makes each feature map contribute differently to target segmentation. Based on the above, adaptive feature pooling is used for feature fusion;
[0056] S23, region of interest acquisition. In the detection head, a series of convolution operations are performed on the input feature map to obtain the target category information and mask information. This image uses different colors to represent different temperatures. The integration is a pseudo-color map of the surface temperature distribution of the detected crop and its environment. If the crop environment temperature is in good condition, the detected infrared image is relatively uniform. Otherwise, the infrared image will present an uneven color picture. Specifically:
[0057] Each ROI (Region of The feature map Φ obtained after fusion will be used as the input feature map to input the detection head to output the result. Regarding the generation of the target mask, the feature map Φ needs to be convolved three times to obtain Φ'. Then, it will enter two branches, one of which uses FCN (Fully-connected Network) to generate the corresponding segmentation mask M for each category; the other branch inputs the feature map Φ' into two fully connected layers, and outputs a feature vector V containing the foreground and background binary classification information of each point in the feature map. Then, the feature vector V and the segmentation mask M are resized to make them the same size as Φ'. The resize correction operation of M is performed using deconvolution upsampling, and the feature map with more foreground and background pixel-level details is fused with the mask obtained by FCN to obtain a high-precision segmentation mask of the final target. After segmentation, the object to be tested can be extracted. The infrared image carries the object temperature information, which can better extract the temperature detection information corresponding to the segmented target. At the same time, the coordinate position of the segmented crop can be better obtained for subsequent precise fixed-point control;
[0058] S24, obtaining detection head results.
[0059] In one embodiment of the present invention, S3 includes:
[0060] S31. Dapeng greenhouse provides a suitable environment for crops by passively absorbing solar radiation. The indoor environment is dynamically affected by the state of the control equipment, the growth of crops and the disturbance variables of the outdoor environment. In addition, the heat exchange and heat dissipation rates of the greenhouse are closely related to structural parameters such as lighting materials, area, and geographical location. The greenhouse internal environment mechanism model established based on the energy and material balance theory can describe the law of change of the system and is suitable for the optimal design of environmental set values in greenhouse production. First, the temperature change that will occur in the crop when it is not affected by the controlled equipment is calculated through the dynamic differential equation of the greenhouse environment and the type of crop.
[0061] The dynamic differential equation of the greenhouse environment:
[0062]
[0063] Where X m is the m-dimensional indoor environmental variable (temperature, humidity, CO2 concentration, etc.); U n is the state variable of the n-dimensional greenhouse control equipment (heating, dehumidification, supplementary lighting, etc.); V o is the o-dimensional crop physiological state variable (photosynthetic rate, respiration rate, transpiration rate, etc.); D p is the p-dimensional outdoor environmental disturbance variable (temperature, light, wind speed, etc.); C q are the structural parameters of the q-dimensional greenhouse (size, material, geographical location, etc.); t is time, s; f is a nonlinear function based on energy and material balance.
[0064] S32, after the change in the ambient temperature in the greenhouse is fed back to the PID control module through the processor, the environmental control method is gradually transformed into a dynamic control method based on real-time feedback from environmental monitoring and taking the set value as the target, in view of the time-varying, time-delayed and nonlinear characteristics of the greenhouse system. PID control is a classic feedback control method that uses proportional, integral and environmental dynamic differential equation operations to feedback control the system error, thereby ensuring stable operation of the system:
[0065] The output u(t) of the PID control for regional regulation is:
[0066]
[0067] Where e(t) is the difference between the actual measured value and the set value, K p , T1, T D The PID controller adjusts the three parameters K p , T1, T D It can improve the dynamic and static characteristics of the system, with simple principle and easy use.
[0068] Then, based on the target area points obtained from the previous infrared images, smart devices are used to perform small-scale temperature control to bring the target area to the ideal growth temperature range.
[0069] S33. After the first round of temperature information acquisition and regulation, the sensor group and infrared imager work continuously and monitor in real time. When the first round of adjustment fails to achieve the ideal goal, a second or even more rounds of adjustment will be carried out to ensure that crops grow in a relatively ideal temperature range.
[0070] In the present invention, in order to better combine the surrounding environment to achieve precise control, the influence of the dynamic differential equation of the regional environment should be considered in the control output of PID. With the differential time constant T D Multiply and combine the result to get the output u(t) for regional regulation. This will be more accurate than the overall regulation of the greenhouse. The environmental dynamic differential equation is affected by factors such as the geographical environment. In different areas of the same greenhouse, the environmental dynamic differential equations of each area are also different. Therefore, the regulation of a single area will be more accurate and economical than the unified regulation of the entire greenhouse. For example: if the greenhouse is divided into 8x8 grid areas, a small amount of precise regulation for a single area can effectively save one-sixty-fourth of the original cost, and will not affect the growth of crops in other normal areas.
[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for accurately adjusting the growth environment of greenhouse crops combined with the PANet algorithm, characterized in that: The following steps are included: S1. Arrange at least one set of temperature sensors at different locations in the greenhouse to sense the greenhouse temperature, and use DTU to upload the data and compare it with the ideal temperature range for crop growth; S2, using an infrared temperature detector to extract infrared images of greenhouse crops, segmenting the extracted infrared images of greenhouse crops through the PANet algorithm, dividing the greenhouse into several areas, and finally obtaining mask information including position and edge information; S3. Calculate the change in greenhouse ambient temperature through the environmental dynamic differential equation, and combine it with the feedback control method of the PID controller to regulate the greenhouse area that needs to be regulated, and accurately adjust the temperature of a single target area. After the adjustment is completed, use regional image segmentation and temperature sensor to sense again to obtain data feedback to observe whether secondary regulation is needed to achieve stable control.
2. According to claim 1, a method for accurately adjusting the growth environment of greenhouse crops combined with the PANet algorithm is characterized in that: The S1 includes: S11, firstly, evenly distribute the temperature sensor groups at the same interval positions in the greenhouse; S12, multiple sensors collect temperature data from different planting sites simultaneously but the temperature data transmission intervals are different; S13, all sensor groups are connected to a DTU respectively to upload the collected temperature data to the IoT cloud platform; S14. Through the processor, the temperature data obtained by the IoT cloud platform is compared with the ideal growth temperature range of the crop to check whether the environment in which the crop is located is conducive to its growth. The sensors in the same sensor group are numbered, and the upload delay of the front and rear sensors is set, and the upload frequency of the front and rear sensors is the same.
3. According to claim 1, a method for accurately adjusting the growth environment of greenhouse crops combined with the PANet algorithm is characterized in that: The S2 includes: S21, feature extraction, after obtaining a pre-processed image, PANet will make ResNet perform preliminary multi-scale feature extraction; S22, feature fusion, then use top-down and bottom-up feature fusion methods to fuse the multi-scale feature maps, and then use adaptive feature pooling to downsample the obtained region proposals for input into the detection head; S23, obtaining the region of interest. In the detection head, a series of convolution operations are performed on the input feature map to obtain the target category information and mask information. This image uses different colors to represent different temperatures. The integration together is a pseudo-color map of the surface temperature distribution of the detected crop and its environment. If the crop environment temperature is in good condition, the detected infrared image is relatively uniform. Otherwise, the infrared image will present an uneven color picture. S24, obtaining detection head results.
4. According to claim 3, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: In the S21 feature extraction, after obtaining a pre-processed image, PANet causes ResNet to perform preliminary multi-scale feature extraction and obtain multiple multi-scale feature maps.
5. According to claim 4, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: In the S21 feature fusion, the feature map extracted by ResNet is used to construct a feature pyramid, and an additional bottom-up feature enhancement path is constructed for the feature pyramid. In this path, the feature map is fused from bottom to top, from large size to small size, so that the high-level feature map obtains detailed information on the original basis. PANet uses a shallower N i And deeper P i+1 The fusion method gets its next layer N i+1 , it first passes a 3*3 convolution with a stride of 2 on N i Downsample, and then add P i+1 The feature map after downsampling is then fused, and then a 3*3 convolution is used to fuse the features to increase the representation ability of the fused features, and finally the ReLU activation function is used to nonlinearize the features; After obtaining the multi-level feature map N, RPN is used to classify the foreground and background in the feature map. RPN will generate a series of anchor frames in each level of the feature map, and score and fine-tune the size of each anchor frame through regression and classification. Finally, a set of anchor frame coordinates will be output for each feature map for subsequent operations. However, the scales of the feature maps at each level are different, and the global and feature information they contain are different, which makes each feature map contribute differently to the target segmentation. Based on the above, adaptive feature pooling is used for feature fusion.
6. According to claim 5, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: S23, the area of interest is obtained specifically as follows: The feature map Φ obtained after each ROI fusion will be input into the detection head as the input feature map to output the result. Regarding the generation of the target mask, the feature map Φ needs to be convolved three times to obtain Φ'. Then, it will enter two branches, one of which uses FCN to generate the corresponding segmentation mask M for each category; the other branch inputs the feature map Φ' into two fully connected layers, and outputs a feature vector V containing the foreground and background binary classification information of each point in the feature map. Then, the feature vector V and the segmentation mask M are resized to make them the same size as Φ'. The resize correction operation of M is performed using deconvolution upsampling, and the feature map with more pixel-level details of the foreground and background is fused with the mask obtained by FCN to obtain a high-precision segmentation mask of the final target. After segmentation, the object to be tested is extracted. The infrared image carries the object temperature information, and the temperature detection information corresponding to the segmented target can be extracted. At the same time, the coordinate position of the segmented crop can be obtained for subsequent precise fixed-point control.
7. According to claim 1, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: In S3, the change in the temperature of the greenhouse environment is calculated by using the dynamic differential equation of the environment. The dynamic differential equation of the greenhouse environment is: Where X m is the m-dimensional indoor environment variable; U n is the n-dimensional greenhouse control equipment state variable; V o is o-dimensional crop physiological state variable; D p is the p-dimensional outdoor environment disturbance variable; C q is the q-dimensional greenhouse structure parameter; t is time, s; f is a nonlinear function based on energy and material balance.
8. According to claim 7, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: In S3, the feedback control method of the PID controller is combined to control the greenhouse area that needs to be regulated, and the temperature of a single target area is accurately adjusted, including: After the change in the ambient temperature in the greenhouse is fed back to the PID control module through the processor, the environmental control method is gradually transformed into a dynamic control method based on the set value and real-time feedback from environmental monitoring. The system error is feedback controlled through proportional, integral and combined environmental dynamic differential equation operations: The output u(t) of the PID control for regional regulation is: Where e(t) is the difference between the actual measured value and the set value, K p , T1, T D They are proportional gain, integral time constant and differential time constant respectively. The PID controller adjusts the three parameters K p , T1, T D That is, it can improve the dynamic and static characteristics of the system, and then use smart devices to perform small-scale temperature control based on the target area points obtained from the previous infrared images to make the target area reach the ideal growth temperature range.
9. According to claim 8, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: The environmental control method is specifically as follows: Combined with the dynamic differential equation of the greenhouse environment, PID control can then ensure the stable control of the greenhouse environment through feedback adjustment. The average standard error of the adaptive control strategy is reduced. The system selects the model category according to the input and output data, and uses system identification to adaptively adjust the unknown parameters in the model to ensure that the actual temperature gradually approaches the set value.
10. According to claim 1, a method for accurately adjusting the greenhouse crop growth environment in combination with the PANet algorithm is characterized in that: Applicable to the regulation system consisting of the following modules: The ambient temperature sensing module includes at least one set of temperature sensors, which are arranged at different locations in the greenhouse to sense the greenhouse temperature at different locations and at different time points. The infrared temperature tester is set up in the greenhouse to obtain infrared images of crops. Data transmission module, using DTU to transmit temperature data and infrared images to the cloud platform, and connect relays for remote control; The control and adjustment module segments the extracted infrared image of greenhouse crops through the PANet algorithm, divides the greenhouse into several areas, and finally obtains mask information including position and edge information; calculates the change in greenhouse ambient temperature through the environmental dynamic differential equation, and combines the feedback control method of the PID controller to regulate the greenhouse area that needs to be regulated, and accurately adjusts the temperature of a single target area. After the adjustment is completed, the regional image segmentation and temperature sensor are used to perceive again to obtain data feedback to observe whether secondary regulation is needed to achieve stable control.
