Ploughing intelligent protection monitoring system and method based on image recognition

Abstract

The invention provides a cultivated land intelligent protection monitoring system and method based on image recognition, and belongs to the technical field of cultivated land intelligent protection monitoring. The system comprises an image acquisition module, an image processing module, a data analysis module and an alarm module; the image acquisition module is used for acquiring image data of cultivated land in real time; the image processing module is used for performing feature extraction and processing on the acquired image data to obtain a clean image of the cultivated land; the data analysis module is used for carrying out image recognition and analysis on the clean image and judging whether the cultivated land is abnormal or not; and the alarm module is used for sending out alarm information in time when an abnormal condition is detected. The problems that an existing farmland monitoring system based on image recognition still has defects in the aspects of recognition precision, real-time performance, intelligent degree and the like, and actual requirements of farmland protection are difficult to meet are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent farmland protection and monitoring, and specifically relates to an intelligent farmland protection and monitoring system and method based on image recognition. Background Art

[0002] With the advancement of agricultural modernization, farmland protection has become an important part of agricultural production. Traditional methods of farmland monitoring mainly rely on manual inspections and field observations, which have problems such as low efficiency, high cost, and easy omissions. Intelligent farmland protection monitoring is a new technical means that has emerged in the field of farmland protection in recent years. It uses modern information technology, such as video monitoring, remote sensing monitoring, and big data analysis, to monitor and warn farmland in real time. The emergence of this technology is to cope with the severe challenges currently facing farmland protection, such as the frequent occurrence of "non-agriculturalization" and "non-grainization" of farmland, and the lack of implementation of farmland protection responsibilities. In recent years, with the continuous development of image recognition technology, its application in the agricultural field has gradually increased, providing new technical means for farmland monitoring.

[0003] The existing farmland monitoring system based on image recognition still has shortcomings in recognition accuracy, real-time performance, and intelligence, and it is difficult to meet the actual needs of farmland protection. Summary of the invention

[0004] The present invention provides a farmland intelligent protection monitoring system and method based on image recognition, which aims to solve the problem that the existing farmland monitoring system based on image recognition still has deficiencies in recognition accuracy, real-time performance, intelligence level, etc., and is difficult to meet the actual needs of farmland protection.

[0005] The embodiment of the present invention provides a farmland intelligent protection monitoring system based on image recognition, including an image acquisition module, an image processing module, a data analysis module and an alarm module;

[0006] The image acquisition module is used to collect image data of cultivated land in real time;

[0007] The image processing module is used to extract and process the collected image data to obtain a clean image of the cultivated land;

[0008] The data analysis module is used to perform image recognition and analysis on clean images to determine whether there are abnormal conditions in the cultivated land;

[0009] The alarm module is used to send out alarm information in time when an abnormal situation is detected.

[0010] Furthermore, the image acquisition module includes a high-definition camera and an image transmission device, wherein the high-definition camera is used to capture real-time images of cultivated land, and the image transmission device is used to transmit the captured image data to the image processing module.

[0011] Furthermore, the image processing module includes a feature extraction unit and a clean image generation unit, wherein the clean image generation unit uses the image features extracted by the feature extraction unit for learning and outputs a denoised clean image;

[0012] The feature extraction unit is used to extract feature information of cultivated land in the collected image, and the feature information includes color, texture, and shape;

[0013] The clean image generation unit is used to perform denoising preprocessing on the collected image data.

[0014] Furthermore, the feature extraction unit includes four convolutional layers of a convolutional neural network and a spatial channel attention mechanism;

[0015] The spatial channel attention mechanism includes two sub-modules, the channel attention mechanism and the spatial attention mechanism. The combination of the two sub-modules can reduce the number of parameters and the amount of calculation, adaptively learn the channel and spatial attention weights, multiply the output feature elements of the channel attention mechanism and the spatial attention mechanism element by element, constrain and enhance the features, so that the feature expression ability of the convolutional neural network is improved. The extracted features are:

[0016] B' conv (F) = R{e 3×3 [R(e 3×3 (e 1×1 (E))))))]} (1)

[0017] N' c (F) = N c [B' conv (E)] (2)

[0018] B' out (E)=M c [B' out (E)]×N s [N' c (E)] (3);

[0019] Among them, B' conv (F) represents the output features after four convolutional layers and ReLU activation function, N' c (E) represents the image features extracted after the channel attention mechanism, B' out (E) represents the image features extracted after the feature extraction unit, F represents the input image features, R represents the ReLU activation function, e 3×3 With e 1×1 Respectively represent the convolutional layers with kernel sizes of 3×3 and 1×1, N c Represents the weight coefficient of the channel attention mechanism, Ns Represents the weight coefficient of the spatial attention mechanism, and × represents weighted multiplication.

[0020] Furthermore, the channel attention mechanism takes the extracted feature E as input, performs maximum pooling and global average pooling on the feature F in the spatial dimension, compresses the size, and facilitates learning channel features. Then, the results of maximum pooling and global average pooling are respectively sent to a multi-layer perceptron for learning, and a feature map is obtained. Then, the results output by the multi-layer perceptron are weighted, and then processed by the mapping of the Sigmoid activation function, and finally the channel attention weight matrix coefficient M is obtained. c ;

[0021] The channel attention weight matrix N c (E) is expressed as:

[0022] N c (E)=ω(MLP(AvgPool(E))+MLP(MaxPool(E))))

[0023] =ω(P 1 (P 0 (E C avg ))+P 1 (P 0 (E C max ))) (4)

[0024] Among them, E C avg represents the channel feature after average pooling, E C max represents the channel feature after maximum pooling, ω represents the activation function Sigmoid, P 1 Represents the parameters in the shared multilayer perceptron, used to share the input P 0 The ReLU activation function in .

[0025] Furthermore, the spatial attention mechanism performs global maximum pooling and global average pooling on the input feature map E in the channel dimension, and then concatenates the results of the global maximum pooling and global average pooling according to the channel, and then convolves the concatenated results through a convolutional neural network. Finally, the spatial attention weight matrix coefficient N is obtained through the Sigmoid activation function. s ;

[0026] The spatial attention weight matrix N s (E) is expressed as:

[0027] N c (E) = ω{e 7×7[(AvgPool(E); MaxPool(E)]}

[0028] =ω[e 7×7 (E S avg ; E S max ) ] (5)

[0029] Among them, E S avg represents the spatial features obtained after average pooling, E S max represents the spatial features obtained after maximum pooling, e 7×7 Indicates that the kernel size of the convolution layer is 7×7.

[0030] Further, the data analysis module includes an image recognition unit and an abnormality judgment unit;

[0031] The image recognition unit is used to perform image recognition on the clean image;

[0032] The abnormality judgment unit is used to judge whether there is an abnormality in the cultivated land according to the recognition result.

[0033] Further, the alarm module includes an alarm information generating unit and an alarm information sending unit;

[0034] The alarm information generating unit is used to generate corresponding alarm information according to the type and degree of the abnormality when an abnormality is detected;

[0035] The alarm information sending unit is used to send the generated alarm information to relevant personnel or equipment.

[0036] A method for intelligent farmland protection and monitoring based on image recognition, comprising the above-mentioned intelligent farmland protection and monitoring system based on image recognition, characterized in that it also includes the following steps:

[0037] Step S1: collecting image data of cultivated land in real time through an image acquisition module;

[0038] Step S2: extracting and processing the collected image data through an image processing module to obtain a clean image of the cultivated land;

[0039] The feature extraction includes extracting feature information of color, texture and shape of cultivated land from the clean image;

[0040] The image processing includes denoising processing;

[0041] Step S3: Perform image recognition and analysis on the clean image to determine whether there is any abnormality in the cultivated land;

[0042] The image recognition includes identifying crops, soil, and buildings in cultivated land;

[0043] The abnormality judgment includes judging whether there are abnormal conditions such as crop pests and diseases, soil pollution, and illegal occupation in the cultivated land according to the recognition results.

[0044] Step S4: When an abnormal situation is detected, an alarm message is issued in time;

[0045] The alarm information includes information on the abnormality type, abnormality degree, and abnormality location.

[0046] Furthermore, the denoising process includes extracting features from the image by a feature extraction unit, and then generating a clean image from the image after feature extraction by a clean image generation unit.

[0047] The beneficial effects of the present invention are:

[0048] The present invention performs image denoising processing through convolutional neural network, which can not only generate clean images, but also avoid the loss of image details, thereby improving the accuracy and efficiency of cultivated land monitoring, and can timely discover and deal with abnormal conditions in cultivated land, and protect cultivated land resources; it realizes the intelligence and automation of cultivated land monitoring, and reduces the cost and labor intensity of manual inspections; it provides rich alarm information, which helps relevant personnel or equipment to take timely measures to deal with it, and improves the timeliness and effectiveness of cultivated land protection.

[0049] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0051] Figure 1 A schematic diagram of the structure of a monitoring system according to an embodiment of the present invention;

[0052] Figure 2 Schematic diagram of the monitoring method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the technical solution of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described in conjunction with the drawings of specific embodiments of the present invention. The same figure marks in the drawings represent the same parts. It should be noted that the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the described 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.

[0054] Reference Figure 1 , an embodiment of the present invention proposes a farmland intelligent protection monitoring system based on image recognition, including an image acquisition module, an image processing module, a data analysis module and an alarm module;

[0055] The image acquisition module is used to collect image data of cultivated land in real time;

[0056] The image processing module is used to extract and process the collected image data to obtain a clean image of the cultivated land;

[0057] The data analysis module is used to perform image recognition and analysis on clean images to determine whether there are abnormal conditions in the cultivated land;

[0058] The alarm module is used to send out alarm information in time when an abnormal situation is detected.

[0059] Furthermore, the image acquisition module includes a high-definition camera and an image transmission device, wherein the high-definition camera is used to capture real-time images of cultivated land, and the image transmission device is used to transmit the captured image data to the image processing module.

[0060] Furthermore, the image processing module includes a feature extraction unit and a clean image generation unit, wherein the clean image generation unit uses the image features extracted by the feature extraction unit for learning and outputs a denoised clean image;

[0061] The feature extraction unit is used to extract feature information of cultivated land in the collected image, and the feature information includes color, texture, and shape;

[0062] The clean image generation unit is used to perform denoising preprocessing on the collected image data to improve the image quality.

[0063] The feature extraction unit includes four convolutional layers of a convolutional neural network and a spatial channel attention mechanism;

[0064] While enhancing the extraction of image features, the model's attention is paid to the image features, which are then input into a clean image generation unit without jump connections and batch normalization layers to fully extract the clean image signal in the image and generate a denoised clean image.

[0065] The spatial channel attention mechanism includes two sub-modules, the channel attention mechanism and the spatial attention mechanism. The combination of the two sub-modules can reduce the number of parameters and the amount of calculation, adaptively learn the channel and spatial attention weights, multiply the output feature elements of the channel attention mechanism and the spatial attention mechanism element by element, constrain and enhance the features, so that the feature expression ability of the convolutional neural network is improved. The extracted features are:

[0066] B' conv (F) = R{e 3×3 [R(e 3×3 (e 1×1 (E))))))]} (1)

[0067] N' c (F) = N c [B' conv (E)] (2)

[0068] B' out (E)=M c [B' out (E)]×N s [N' c (E)] (3);

[0069] Among them, B' conv (F) represents the output features after four convolutional layers and ReLU activation function, N' c (E) represents the image features extracted after the channel attention mechanism, B' out (E) represents the image features extracted after the feature extraction unit, F represents the input image features, R represents the ReLU activation function, e 3×3 With e 1×1 Respectively represent the convolutional layers with kernel sizes of 3×3 and 1×1, N c Represents the weight coefficient of the channel attention mechanism, N s Represents the weight coefficient of the spatial attention mechanism, and × represents weighted multiplication.

[0070] The channel attention mechanism takes the extracted feature E as input, performs maximum pooling and global average pooling on the feature F in the spatial dimension, compresses the size, and facilitates learning channel features. Then, the results of maximum pooling and global average pooling are respectively sent to a multi-layer perceptron for learning, and a feature map is obtained. The results of the multi-layer perceptron output are weighted, and then mapped by the Sigmoid activation function to finally obtain the channel attention weight matrix coefficient M c ;

[0071] The channel attention weight matrix N c (E) is expressed as:

[0072] N c (E)=ω(MLP(AvgPool(E))+MLP(MaxPool(E))))

[0073] =ω(P 1 (P 0 (E C avg ))+P 1 (P 0 (E C max ))) (4)

[0074] Among them, E C avg represents the channel feature after average pooling, E C max represents the channel feature after maximum pooling, ω represents the activation function Sigmoid, P 1 Represents the parameters in the shared multilayer perceptron, used to share the input P 0 The ReLU activation function in .

[0075] The spatial attention mechanism performs global maximum pooling and global average pooling on the input feature map E in the channel dimension, then concatenates the results of global maximum pooling and global average pooling according to the channel, and then convolves the concatenated results through a convolutional neural network. Finally, the spatial attention weight matrix coefficient N is obtained through the Sigmoid activation function. s ;

[0076] The spatial attention weight matrix N s (E) is expressed as:

[0077] N c (E) = ω{e 7×7 [(AvgPool(E); MaxPool(E)]}

[0078] =ω[e 7×7 (E Savg ; E S max ) ] (5)

[0079] Among them, E S avg represents the spatial features obtained after average pooling, E S max represents the spatial features obtained after maximum pooling, e 7×7 Indicates that the kernel size of the convolution layer is 7×7.

[0080] Further, the data analysis module includes an image recognition unit and an abnormality judgment unit;

[0081] The image recognition unit is used to perform image recognition on the clean image to identify objects such as crops, soil, buildings, etc. in the cultivated land;

[0082] The abnormality judgment unit is used to judge whether there are abnormal conditions in the cultivated land, such as crop pests and diseases, soil pollution, illegal occupation, etc., according to the recognition results.

[0083] Further, the alarm module includes an alarm information generating unit and an alarm information sending unit;

[0084] The alarm information generating unit is used to generate corresponding alarm information according to the type and degree of the abnormality when an abnormality is detected;

[0085] The alarm information sending unit is used to send the generated alarm information to relevant personnel or equipment so that timely measures can be taken to handle it.

[0086] Reference Figure 2 , a method for intelligent protection and monitoring of cultivated land based on image recognition, comprising the following steps:

[0087] Step S1: collecting image data of cultivated land in real time through an image acquisition module;

[0088] Step S2: extracting and processing the collected image data through an image processing module to obtain a clean image of the cultivated land;

[0089] The feature extraction includes extracting feature information of color, texture and shape of cultivated land from the clean image;

[0090] The image processing includes denoising to improve image quality;

[0091] Step S3: Perform image recognition and analysis on the clean image to determine whether there is any abnormality in the cultivated land;

[0092] The image recognition includes identifying crops, soil, and buildings in cultivated land;

[0093] The abnormality judgment includes judging whether there are abnormal conditions such as crop pests and diseases, soil pollution, and illegal occupation in the cultivated land according to the recognition results.

[0094] Step S4: When an abnormal situation is detected, an alarm message is issued in time;

[0095] The alarm information includes information on the abnormality type, abnormality degree, and abnormality location, so that relevant personnel or equipment can take timely measures to handle the problem.

[0096] Furthermore, the denoising process includes extracting features from the image by a feature extraction unit, and then generating a clean image from the image after feature extraction by a clean image generation unit.

[0097] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A farmland intelligent protection monitoring system based on image recognition, characterized in that: It includes image acquisition module, image processing module, data analysis module and alarm module; The image acquisition module is used to collect image data of cultivated land in real time; The image processing module is used to extract and process the collected image data to obtain a clean image of the cultivated land; The data analysis module is used to perform image recognition and analysis on clean images to determine whether there are abnormal conditions in the cultivated land; The alarm module is used to send out alarm information in time when an abnormal situation is detected.

2. According to claim 1, the intelligent farmland protection monitoring system based on image recognition is characterized in that: The image acquisition module includes a high-definition camera and an image transmission device. The high-definition camera is used to capture real-time images of cultivated land, and the image transmission device is used to transmit the captured image data to the image processing module.

3. The intelligent farmland protection monitoring system based on image recognition according to claim 1 is characterized in that: The image processing module includes a feature extraction unit and a clean image generation unit, wherein the clean image generation unit uses the image features extracted by the feature extraction unit for learning and outputs a denoised clean image; The feature extraction unit is used to extract feature information of cultivated land in the collected image, and the feature information includes color, texture, and shape; The clean image generation unit is used to perform denoising preprocessing on the collected image data.

4. The intelligent farmland protection monitoring system based on image recognition according to claim 1 is characterized in that: The feature extraction unit includes four convolutional layers of a convolutional neural network and a spatial channel attention mechanism; The spatial channel attention mechanism includes two sub-modules, the channel attention mechanism and the spatial attention mechanism. The combination of the two sub-modules can reduce the number of parameters and the amount of calculation, adaptively learn the channel and spatial attention weights, multiply the output feature elements of the channel attention mechanism and the spatial attention mechanism element by element, constrain and enhance the features, so that the feature expression ability of the convolutional neural network is improved. The extracted features are: B' conv (F)=R{e 3×3 [R(e 3×3 (e 1×1 (E)))))]} (1) N' c (F)=N c [B' conv (E)] (2) B' out (E)=M c [B' out (E)]×N s [N' c (E)] (3); Among them, B' conv (F) represents the output features after four convolutional layers and ReLU activation function, N' c (E) represents the image features extracted after the channel attention mechanism, B' out (E) represents the image features extracted after the feature extraction unit, F represents the input image features, R represents the ReLU activation function, e 3×3 With e 1×1 Respectively represent the convolutional layers with kernel sizes of 3×3 and 1×1, N c Represents the weight coefficient of the channel attention mechanism, N s Represents the weight coefficient of the spatial attention mechanism, and × represents weighted multiplication.

5. The intelligent farmland protection monitoring system based on image recognition according to claim 1 is characterized in that: The channel attention mechanism takes the extracted feature E as input, performs maximum pooling and global average pooling on the feature F in the spatial dimension, compresses the size, and facilitates learning channel features. Then, the results of maximum pooling and global average pooling are respectively sent to a multi-layer perceptron for learning, and a feature map is obtained. The results of the multi-layer perceptron output are weighted, and then mapped by the Sigmoid activation function to finally obtain the channel attention weight matrix coefficient M c ; The channel attention weight matrix N c (E) is expressed as: N c (E)=ω(MLP(AvgPool(E))+MLP(MaxPool(E)))) =ω(P1(P0(E C avg ))+P1(P0(E C max ))) (4) Among them, E C avg represents the channel feature after average pooling, E C max represents the channel feature after maximum pooling, ω represents the activation function Sigmoid, and P1 represents the parameters in the shared multi-layer perceptron, which is used to share the ReLU activation function in the input P0.

6. The intelligent farmland protection monitoring system based on image recognition according to claim 1 is characterized in that: The spatial attention mechanism performs global maximum pooling and global average pooling on the input feature map E in the channel dimension, then concatenates the results of global maximum pooling and global average pooling according to the channel, and then convolves the concatenated results through a convolutional neural network. Finally, the spatial attention weight matrix coefficient N is obtained through the Sigmoid activation function. s ; The spatial attention weight matrix N s (E) is expressed as: N c (E)=ω{e 7×7 [(AvgPool(E);MaxPool(E)]} =ω[e 7×7 (AND S avg ;AND S max ) ] (5) Among them, E S avg represents the spatial features obtained after average pooling, E S max represents the spatial features obtained after maximum pooling, e 7×7 Indicates that the convolutional layer kernel size is 7×7.

7. The intelligent farmland protection and monitoring system based on image recognition according to claim 1 is characterized in that: The data analysis module includes an image recognition unit and an abnormality judgment unit; The image recognition unit is used to perform image recognition on the clean image; The abnormality judgment unit is used to judge whether there is an abnormality in the cultivated land according to the recognition result.

8. The intelligent farmland protection and monitoring system based on image recognition according to claim 1 is characterized in that: The alarm module includes an alarm information generating unit and an alarm information sending unit; The alarm information generating unit is used to generate corresponding alarm information according to the type and degree of the abnormality when an abnormality is detected; The alarm information sending unit is used to send the generated alarm information to relevant personnel or equipment.

9. A method for intelligent farmland protection and monitoring based on image recognition, comprising the intelligent farmland protection and monitoring system based on image recognition according to any one of claims 1 to 8, characterized in that: The following steps are also included: Step S1: collecting image data of cultivated land in real time through an image acquisition module; Step S2: extracting and processing the collected image data through an image processing module to obtain a clean image of the cultivated land; The feature extraction includes extracting feature information of color, texture and shape of cultivated land from the clean image; The image processing includes denoising processing; Step S3: Perform image recognition and analysis on the clean image to determine whether there is any abnormality in the cultivated land; The image recognition includes identifying crops, soil, and buildings in cultivated land; The abnormality judgment includes judging whether there are abnormal conditions such as crop pests, soil pollution, and illegal occupation in the cultivated land according to the recognition results; Step S4: When an abnormal situation is detected, an alarm message is issued in time; The alarm information includes information on the abnormality type, abnormality degree, and abnormality location.

10. The method for intelligent farmland protection and monitoring based on image recognition according to claim 9, characterized in that: The denoising process includes extracting features from an image through a feature extraction unit, and then generating a clean image from the image after feature extraction through a clean image generation unit.

Images