Tobacco leaf disease identification and control method, device and equipment and storage medium

Abstract

The invention discloses a tobacco leaf disease identification and control method, device and equipment and a storage medium, and relates to the technical field of plant disease identification, and the tobacco leaf disease identification and control method comprises the steps: constructing and training a disease identification model based on physical perception constraint and a hierarchical graph neural network; obtaining a to-be-identified tobacco leaf image, processing the to-be-identified tobacco leaf image through the disease identification model, and generating a disease category prediction result; and generating a disease control scheme according to a preset multi-modal large model and the disease category prediction result. The accuracy of tobacco leaf disease identification can be improved. Therefore, the reliability of a disease control scheme generated by a large model is effectively improved, the tobacco yield is improved, and economic loss is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of plant disease identification, and in particular to a tobacco leaf disease identification and prevention method, device, equipment and storage medium. BACKGROUND

[0002] At present, the existing deep learning-based method can identify specific categories of tobacco leaf diseases, but for the categories with similar appearances of tobacco leaf diseases, different cluttered background interference in the scene, different illumination angle changes and the like, the classification and identification result is not accurate. Therefore, how to improve the accuracy of the identification of tobacco leaf diseases is still a problem to be solved.

[0003] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0004] The main purpose of the present application is to provide a tobacco leaf disease identification and prevention method, device, equipment and storage medium, aiming to solve the technical problem of how to improve the accuracy of the identification of tobacco leaf diseases.

[0005] To achieve the above purpose, the present application provides a tobacco leaf disease identification and prevention method, which comprises: constructing and training a disease identification model based on physical perception constraints and a hierarchical graph neural network; obtaining a tobacco leaf image to be identified, processing the tobacco leaf image to be identified through the disease identification model, and generating a disease category prediction result; generating a disease prevention scheme according to a preset multi-modal large model and the disease category prediction result.

[0006] In an embodiment, the step of constructing and training a disease identification model based on physical perception constraints and a hierarchical graph neural network comprises: obtaining target tobacco leaf training image data; extracting disease features of the target tobacco leaf training image data based on physical perception constraints; constructing a hierarchical graph neural network architecture; training a hybrid loss function of the hierarchical graph neural network architecture through the target tobacco leaf training image data and the disease features, to obtain a disease identification model.

[0007] In an embodiment, the step of obtaining target tobacco leaf training image data comprises: obtaining original tobacco leaf training image data; adjusting the original tobacco leaf training image data according to a preset network structure model requirement to obtain target tobacco leaf training image data.

[0008] In an embodiment, the step of extracting the disease feature of the target tobacco training image data based on the physical perception constraint comprises: The target tobacco training image data is converted in color space to extract a lesion chrominance abnormal feature; The lesion chrominance abnormal feature is enhanced in edge texture by using a morphological gradient operator to obtain an enhanced edge feature; A lesion area proportion constraint equation is established to calculate the area range of the enhanced edge feature; The disease feature is generated according to the lesion chrominance abnormal feature, the enhanced edge feature, and the area range.

[0009] In an embodiment, the step of training the hybrid loss function of the layered graph neural network architecture by the target tobacco training image data and the disease feature comprises: A predicted probability distribution result is obtained by the target tobacco training image data, the disease feature, and the layered graph neural network architecture; A real label distribution is obtained, a difference degree between the predicted probability distribution result and the real label distribution is calculated, and the difference degree is taken as a lesion distribution consistency loss; A boundary adjustment loss is generated according to the adjustment of the boundary distance based on the disease feature; The hybrid loss function is obtained according to the lesion distribution consistency loss and the boundary adjustment loss.

[0010] In an embodiment, the step of processing the to-be-identified tobacco image by the disease identification model to generate a disease category prediction result comprises: The to-be-identified tobacco image is evenly divided into multiple blocks and defined as graph nodes, and a feature vector of the graph nodes is extracted to obtain a node initial feature; The connection weight between nodes is calculated according to the feature similarity and spatial position relationship of the node initial feature, and a normalized adjacency matrix is constructed based on the connection weight; The node initial feature is processed by a graph convolution layer with residual connection to obtain a bottom layer feature representation; A multi-head self-attention mechanism is used to capture the long-range dependency relationship of the bottom layer feature representation to generate a high layer feature; Global information of the high layer feature is aggregated to generate a graph level representation, and a disease category prediction result is generated according to the graph level representation.

[0011] In an embodiment, the step of generating a disease control scheme according to the preset multi-modal large model and the disease category prediction result comprises: The disease category prediction result is analyzed according to the preset multi-modal large model and a preset expert knowledge base to obtain an analysis result; generating a disease control scheme including recommended drugs, use period, dosage, and prevention time according to the analysis result.

[0012] In addition, to achieve the above-mentioned purpose, the application further provides a tobacco disease identification and control device, which comprises: The construction module is configured to construct and train a disease identification model based on physical perception constraints and a hierarchical graph neural network. The prediction module is configured to obtain a tobacco image to be identified, process the tobacco image to be identified by using the disease identification model, and generate a disease category prediction result. The generation module is configured to generate a disease control scheme according to a preset multi-modal large model and the generated disease control scheme.

[0013] In addition, to achieve the above-mentioned purpose, the application further provides a tobacco disease identification and control device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the tobacco disease identification and control method as described above.

[0014] In addition, to achieve the above-mentioned purpose, the application further provides a storage medium, which is a computer-readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the tobacco disease identification and control method as described above.

[0015] In addition, to achieve the above-mentioned purpose, the application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the tobacco disease identification and control method as described above.

[0016] The application provides a tobacco disease identification and control method, which constructs and trains a disease identification model based on physical perception constraints and a hierarchical graph neural network, obtains a tobacco image to be identified, processes the tobacco image to be identified by using the disease identification model, generates a disease category prediction result, and generates a disease control scheme according to a preset multi-modal large model and the disease category prediction result.

[0017] As can be seen from the above, the application uses a deep learning model as a basis, adopts a multi-level feature extraction architecture for fine-grained feature extraction, accurately classifies similar disease categories, and improves the accuracy of tobacco identification. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the application and, together with the specification, serve to explain the principles of the application.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0020] Figure 1 A flowchart provided for the tobacco disease identification and prevention method embodiment one of the present application; Figure 2 A flowchart provided for the tobacco disease identification and prevention method embodiment two of the present application; Figure 3 A module structure diagram of the tobacco disease identification and prevention device of the present application embodiment; Figure 4 A device structure diagram of the hardware running environment involved in the tobacco disease identification and prevention method of the present application embodiment.

[0021] The purpose implementation, functional characteristics and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application, and are not used to limit the present application.

[0023] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings in the specification and specific embodiments.

[0024] The main solution of the present application is to construct and train a disease identification model based on physical perception constraints and hierarchical graph neural network; to obtain a to-be-identified tobacco image, to process the to-be-identified tobacco image through the disease identification model, and to generate a disease category prediction result; and to generate a disease prevention scheme according to a preset multi-modal large model and the disease category prediction result.

[0025] At present, the existing deep learning-based method can identify specific categories of tobacco diseases, but for similar categories of tobacco diseases, different clutter background interference in the scene, different light angle changes, etc., the classification and identification result is not accurate. Therefore, how to improve the accuracy of tobacco disease identification is still a problem to be solved.

[0026] The present application uses a deep learning model as the basis, adopts a multi-level feature extraction architecture for fine-grained feature extraction, accurately classifies similar disease categories, and improves the accuracy of tobacco identification.

[0027] Based on this, the present application provides a tobacco disease identification and prevention method, which is described with reference toFigure 1 , Figure 1 FIG. 1 is a flowchart of a first embodiment of the method for identifying and preventing tobacco diseases.

[0028] In this embodiment, the method for identifying and preventing tobacco diseases includes steps S10-S30: Step S10: constructing and training a disease identification model based on physical perception constraints and a hierarchical graph neural network; It should be noted that the execution subject of the present embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a tobacco disease identification and prevention device, etc. In the following, the tobacco disease identification and prevention device is taken as an example to describe the present embodiment and the following embodiments.

[0029] It should be noted that the physical perception constraint method is used to construct a disease-specific feature learning layer for training pictures, which can improve the model's representation of disease characteristics of samples. The hierarchical graph neural network can model local regions in tobacco images as graph nodes and construct connection relationships between nodes according to disease transmission characteristics.

[0030] Step S20: obtaining a tobacco image to be identified, processing the tobacco image to be identified by the disease identification model, and generating a disease category prediction result; It should be noted that after obtaining the tobacco image to be identified, the picture to be evaluated needs to be scaled to the size required by the disease identification model, and then input into the tobacco disease identification network to obtain the probability value of each category of the picture. The disease category of the picture is obtained by selecting the category with the highest probability.

[0031] In one possible way, the step of processing the tobacco image to be identified by the disease identification model to generate a disease category prediction result includes: dividing the tobacco image to be identified into multiple blocks uniformly and defining them as graph nodes, and extracting feature vectors of the graph nodes to obtain node initial features; calculating the connection weight between nodes according to the feature similarity and spatial position relationship of the node initial features, and constructing a normalized adjacency matrix based on the connection weight; processing the node initial features through a graph convolution layer with residual connection to obtain a bottom layer feature representation; capturing the long-range dependency relationship of the bottom layer feature representation using a multi-head self-attention mechanism to generate a high layer feature; aggregating the global information of the high layer feature to generate a graph level representation, and generating a disease category prediction result according to the graph level representation.

[0032] It can be understood that the preprocessed tobacco image to be identified I∈R H×W×3After being input into the disease identification model, the tobacco leaf image to be identified will be uniformly divided into P×P non-overlapping blocks (default P=16), with each block having a size of [missing information]. Each block is a graph node v i A total of N = P × P nodes are generated, and then block features are extracted. Specifically, a CNN backbone network (such as ResNet-50) can be used to extract the feature vector of each block. (D) f =1024), then define the adjacency matrix A∈R N×N , where element A ij Represents node v i With v j The connection strength is determined by both feature similarity and spatial relationship:

[0033] In the formula, Euclidean distance of node feature vectors Feature scale parameter (default) =1.0), Hadamard product (element-by-element multiplication) The spatial similarity matrix is ​​calculated using the following formula:

[0034] in, Indicates the center coordinates of node i. Controlling the spatial decay rate (default σ) p =3.0).

[0035] To avoid gradient explosion, the adjacency matrix A needs to be symmetrically normalized:

[0036] Where I is the identity matrix, Here, D is the normalized adjacency matrix, and D is the degree matrix:

[0037] The graph neural network in the disease identification model of this embodiment adopts a hierarchical structure combining low-level graph convolution and high-level relational inference. The low-level graph convolutional layers (3 layers) adopt an improved GCNII structure, with each layer containing residual connections and initial residual connections. For the node feature matrix X... (0) ∈R N×D (Initial features) are processed, the first Layer computing ( =1, 2, 3):

[0038] wherein, σ is a ReLU activation function, is an initial residual coefficient (default a1=0.2, a2=0.3, a3=0.4), is a weight decay coefficient (default =0.5 / ); ∈R D×D : a learnable parameter matrix.

[0039] The high-level relationship reasoning layer processes the output Z (0) =X (3) ∈R N×D of the bottom layer, and a multi-head self-attention (K heads, default K=8) calculation formula is as follows:

[0040] wherein, t represents a layer number (t=1, 2), represents an output feature of the t-th layer. The attention mechanism is defined as:

[0041] wherein, dk=D / K is a feature dimension of each head, M mask is a spatial mask matrix (prevents non-adjacent node connection), when nodes i, j satisfy ||pi-pj||2>R (default R=3), M mask ij=-∞, otherwise 0.

[0042] Multi-head output splicing and feedforward network:

[0043]

[0044] wherein, the feedforward network FFN is a two-layer MLP:

[0045] Finally, global feature aggregation is performed, and attention pooling is used to generate a graph-level representation:

[0046] wherein, ∈R N×D is a second layer Graph Transformer output, q∈R D is a learnable query vector, ∈R D×D is a learnable parameter.

[0047] Step S30: generating a disease control scheme according to the preset multi-modal large model and the disease class prediction result.

[0048] It can be understood that, according to the existing tobacco disease expert prevention and control knowledge base, the existing multi-modal large model system such as the deepseek model is input; according to the disease prediction result and the expert knowledge base model, the understanding and reasoning ability of the large model is combined to require the model to give specific disease analysis and prevention and control measures.

[0049] In a feasible manner, the step of generating a disease control scheme according to the preset multi-modal large model and the disease category prediction result comprises: analyzing the disease category prediction result according to the preset multi-modal large model and the preset expert knowledge base to obtain an analysis result; and generating a disease control scheme containing recommended drugs, use period, dosage and control time according to the analysis result.

[0050] It can be understood that, the preset prompt words and the disease prediction result can be input into the large model to obtain the disease control scheme. Specific prompt word examples are: according to the provided tobacco disease expert prevention and control knowledge base and model prediction result, please perform detailed analysis and reasoning, and output corresponding detailed prevention and control measures for tobacco diseases, including: disease severity, disease treatment recommended drugs, drug use period and use dosage, etc., and further explain the possible problems and details in the prevention and control.

[0051] The embodiment provides a tobacco disease identification and prevention method. The application constructs and trains a disease identification model based on physical perception constraints and a hierarchical graph neural network; acquires a tobacco image to be identified, processes the tobacco image to be identified through the disease identification model, and generates a disease category prediction result; and generates a disease control scheme according to a preset multi-modal large model and the disease category prediction result.

[0052] As can be seen from the above, the embodiment uses a deep learning model as a basis, adopts a multi-level feature extraction architecture for fine-grained feature extraction, accurately classifies similar disease categories, and improves the accuracy of tobacco identification.

[0053] Based on the first embodiment of the application, in the second embodiment of the application, the same or similar contents as the above embodiment one can refer to the above introduction, and the following will not be repeated. On this basis, please refer to Figure 2 , step S10 further comprises steps S101-S104: Step S101: acquiring target tobacco training image data; It can be understood that, before training, the target tobacco training image data needs to be processed to meet the size requirements of the network structure model, while ensuring the efficiency of training.

[0054] In a possible implementation, the step of obtaining the target tobacco leaf training image data comprises: obtaining original tobacco leaf training image data; and adjusting the original tobacco leaf training image data according to a preset network structure model requirement to obtain the target tobacco leaf training image data.

[0055] It should be noted that the target tobacco leaf training image data comprises pictures and category labels of corresponding diseases. First, the training data is processed to meet the size requirement of the network model. Specifically, the pictures are first extracted by using a deep network model ResNet50 to obtain a feature dimension of 2048, and then disease features are extracted by using a physical perception constraint method to improve the learning efficiency of the model. The data labels are 21 common tobacco disease categories.

[0056] Step S102: extracting disease features of the target tobacco leaf training image data based on physical perception constraints; It can be understood that the physical perception constraint method is used to construct a disease-specific feature learning layer for the training pictures, which can improve the disease feature representation of the model for the samples.

[0057] In a possible implementation, the step of extracting the disease features of the target tobacco leaf training image data based on physical perception constraints comprises: performing color space conversion on the target tobacco leaf training image data to extract lesion chrominance abnormal features; using a morphological gradient operator to strengthen the edge texture of the lesion chrominance abnormal features to obtain strengthened edge features; establishing a lesion area proportion constraint equation to calculate the area range of the strengthened edge features; and generating disease features according to the lesion chrominance abnormal features, the strengthened edge features and the area range.

[0058] It should be noted that the physical perception constraint step specifically comprises: first, using a deep residual network ResNet-50 as a backbone network, and adding a disease-specific feature learning layer on the basis of ImageNet pre-training; then extracting lesion chrominance abnormal features by using HSV color space conversion; then using a morphological gradient operator to strengthen lesion edge texture features; and finally establishing a lesion area proportion constraint equation:

[0059] wherein M(x, y) is a lesion area mask, W x H is an image size, x and y represent horizontal and vertical coordinates of an image pixel respectively, represents a lesion range.

[0060] Step S103: constructing a hierarchical graph neural network architecture; It should be noted that the hierarchical graph neural network architecture can model local regions in a tobacco leaf image as graph nodes, and construct connection relationships between nodes according to disease transmission characteristics.

[0061] Step S104: training the hybrid loss function of the layered graph neural network architecture through the target tobacco training image data and the disease feature, to obtain a disease identification model.

[0062] It should be noted that by training the hybrid loss function, the learning weight of different difficulty samples in the training process can be adjusted, so that the entire model converges faster and better tobacco disease identification results are obtained.

[0063] In a feasible manner, the step of training the hybrid loss function of the layered graph neural network architecture through the target tobacco training image data and the disease feature comprises: obtaining a predicted probability distribution result through the target tobacco training image data, the disease feature and the layered graph neural network architecture; obtaining a real label distribution, calculating the difference degree of the predicted probability distribution result and the real label distribution, and taking the difference degree as a lesion distribution consistency loss; adjusting the boundary distance according to the disease feature to generate a boundary adjustment loss; and obtaining a hybrid loss function according to the lesion distribution consistency loss and the boundary adjustment loss.

[0064] It should be noted that the embodiment adopts the following loss function based on lesion distribution consistency and dynamic boundary adjustment mechanism, wherein the lesion distribution consistency in the loss function takes the KL divergence distribution value of the model prediction result p pred and the real label p real as the first term, and additionally takes the area difference degree of the model prediction area M pred and the real lesion area M gt as the second term, and adds a proportional weight β for adjustment; at the same time, for the dynamic boundary adjustment mechanism in the loss function, the boundary distance is dynamically adjusted according to the classification difficulty of different categories, and the loss function is defined as follows: The lesion distribution consistency loss L phy is:

[0065] Wherein β=0.7 is a balance coefficient.

[0066] The dynamic boundary adjustment mechanism L margin is:

[0067] Wherein δc is automatically adjusted according to the category difficulty, and the initial value m=0.5, d irepresents the distance of the feature vector of sample i to its class center. This distance is usually the Euclidean distance or cosine distance calculated in the feature space. In a classification task, it is often desired that the feature vectors of samples in the same class are clustered around the class center, while the centers of different classes maintain a certain interval.

[0068] The goal of this loss function is to make the feature of each class as compact as possible (intra-class distance is small), and maintain at least m + δc interval between different classes (inter-class distance is large). By dynamically adjusting δc , we can impose greater constraints on difficult-to-classify classes, thereby improving the model's ability to recognize difficult-to-classify classes.

[0069] The embodiment obtains target tobacco training image data; extracts disease feature of the target tobacco training image data based on physical perception constraint; constructs a hierarchical graph neural network architecture; trains a hybrid loss function of the hierarchical graph neural network architecture through the target tobacco training image data and the disease feature, and obtains a disease identification model. The embodiment constructs a disease identification model based on physical perception constraint perception, which can accurately extract diseases in tobacco images and improve the accuracy of tobacco disease identification.

[0070] The application also provides a tobacco disease identification and prevention device, please refer to Figure 3 , the tobacco disease identification and prevention device comprises: a construction module 10 configured to construct and train a disease identification model based on physical perception constraint and hierarchical graph neural network; a prediction module 20 configured to obtain a tobacco image to be identified, process the tobacco image to be identified through the disease identification model, and generate a disease category prediction result; a generation module 30 configured to generate a disease prevention scheme according to a preset multi-modal large model and the disease category prediction result.

[0071] The embodiment provides a tobacco disease identification and prevention method. The application constructs and trains a disease identification model based on physical perception constraint and hierarchical graph neural network; obtains a tobacco image to be identified, processes the tobacco image to be identified through the disease identification model, and generates a disease category prediction result; and generates a disease prevention scheme according to a preset multi-modal large model and the disease category prediction result.

[0072] As can be seen from the above, the embodiment uses a deep learning model as a basis, adopts a multi-level feature extraction architecture for fine-grained feature extraction, accurately classifies similar disease categories, and improves the accuracy of tobacco identification.

[0073] In an embodiment, the construction module 10 is further configured to obtain target tobacco training image data, extract disease characteristics of the target tobacco training image data based on physical perception constraints, construct a hierarchical graph neural network architecture, and train a hybrid loss function of the hierarchical graph neural network architecture through the target tobacco training image data and the disease characteristics to obtain a disease identification model.

[0074] In an embodiment, the construction module 10 is further configured to obtain original tobacco training image data, and adjust the original tobacco training image data according to a preset network structure model requirement to obtain target tobacco training image data.

[0075] In an embodiment, the construction module 10 is further configured to perform color space conversion on the target tobacco training image data to extract disease spot chrominance abnormality characteristics, strengthen edge texture of the disease spot chrominance abnormality characteristics by using a morphological gradient operator to obtain strengthened edge characteristics, establish a disease spot area proportion constraint equation to calculate a region range of the strengthened edge characteristics, and generate disease characteristics according to the disease spot chrominance abnormality characteristics, the strengthened edge characteristics, and the region range.

[0076] In an embodiment, the construction module 10 is further configured to obtain a prediction probability distribution result through the target tobacco training image data, the disease characteristics, and the hierarchical graph neural network architecture, obtain a real label distribution, calculate a difference degree between the prediction probability distribution result and the real label distribution, and take the difference degree as a disease spot distribution consistency loss, adjust a boundary distance according to the disease characteristics to generate a boundary adjustment loss, and obtain a hybrid loss function according to the disease spot distribution consistency loss and the boundary adjustment loss.

[0077] In an embodiment, the prediction module 20 is further configured to uniformly divide the tobacco image to be identified into a plurality of blocks and define the blocks as graph nodes, extract feature vectors of the graph nodes to obtain node initial features, calculate connection weights between nodes according to feature similarity and spatial position relationship of the node initial features, and construct a normalized adjacency matrix based on the connection weights. In an embodiment, the prediction module 20 is further configured to process the node initial features through a graph convolution layer with residual connection to obtain bottom layer feature representations, capture long-range dependency relationships of the bottom layer feature representations by using a multi-head self-attention mechanism to generate high layer features, aggregate global information of the high layer features to generate graph level representations, and generate a disease category prediction result according to the graph level representations.

[0078] In an embodiment, the generation module 30 is further configured to analyze the disease category prediction result according to the preset multi-modal large model and a preset expert knowledge base to obtain an analysis result, and generate a disease control scheme including recommended drugs, use cycles, dosages, and control times according to the analysis result.

[0079] The tobacco disease identification and prevention device provided in the present application adopts the tobacco disease identification and prevention method in the above embodiments, and can solve the technical problem of how to improve the accuracy of tobacco disease identification. Compared with the prior art, the tobacco disease identification and prevention device provided in the present application has the same beneficial effects as the tobacco disease identification and prevention method provided in the above embodiments, and other technical features of the tobacco disease identification and prevention device are the same as the features disclosed in the above embodiments, which will not be repeated here.

[0080] The present application provides a tobacco disease identification and prevention device, which comprises at least one processor and a memory in communication connection with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the tobacco disease identification and prevention method in the above embodiment one.

[0081] Reference will be made to the following description of the embodiments of the present application, taken in conjunction with the accompanying drawings, which illustrate Figure 4 which shows a structural diagram of a tobacco disease identification and prevention device suitable for implementing the embodiments of the present application. The tobacco disease identification and prevention device in the embodiments of the present application can include, but is not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description), PMPs (Portable Media Player), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. Figure 4 The tobacco disease identification and prevention device shown is only an example, and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0082] As Figure 4As shown, the tobacco disease identification and control device can include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to programs stored in a ROM (Read Only Memory) 1002 or programs loaded from a storage device 1003 into a RAM (Random Access Memory) 1004. Various programs and data required for the operation of the tobacco disease identification and control device are also stored in the RAM 1004. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the tobacco disease identification and control device to communicate wirelessly or wired with other devices to exchange data. Although the tobacco disease identification and control device with various systems is shown in the figure, it should be understood that all the shown systems are not required to be implemented or possessed. More or less systems can be alternatively implemented or possessed.

[0083] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are performed.

[0084] The tobacco disease identification and control device provided in the present application adopts the tobacco disease identification and control method in the above-mentioned embodiments, and can solve the technical problem of how to improve the accuracy of the identification of tobacco diseases. Compared with the prior art, the tobacco disease identification and control device provided in the present application has the same beneficial effects as the tobacco disease identification and control method provided in the above-mentioned embodiments, and other technical features in the tobacco disease identification and control device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0085] It should be understood that various aspects of the disclosure can be implemented in hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0086] The above description is merely illustrative of the application and not restrictive.

[0087] The application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e., a computer program) for performing the method for identifying and preventing tobacco diseases in the above embodiments.

[0088] The computer readable storage medium provided by the application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), etc., or any suitable combination thereof.

[0089] The above computer readable storage medium can be included in the tobacco disease identification and prevention device; or can exist separately without being assembled into the tobacco disease identification and prevention device.

[0090] The computer readable storage medium described above carries one or more programs, when the one or more programs are executed by the tobacco disease identification and prevention device, the tobacco disease identification and prevention device: constructs and trains a disease identification model based on physical perception constraints and a hierarchical graph neural network; acquires a to-be-identified tobacco image, processes the to-be-identified tobacco image through the disease identification model, and generates a disease category prediction result; and generates a disease prevention scheme according to a preset multi-modal large model and the disease category prediction result.

[0091] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, Smalltalk, C++, or conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0092] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks noted in succession can in fact be executed substantially concurrently or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowcharts, and combinations thereof, can be implemented by dedicated hardware-based systems that perform the specified functions or operations, or combinations of hardware and software.

[0093] The modules involved in the embodiments of the present application can be implemented in a software manner or in a hardware manner. In some cases, the name of the module does not constitute a limitation on the module itself.

[0094] The readable storage medium provided by the application is a computer readable storage medium, and the computer readable storage medium stores computer readable program instructions (i.e., a computer program) for executing the tobacco disease identification and prevention method described above, and can solve the technical problem of how to improve the accuracy of tobacco disease identification. Compared with the prior art, the computer readable storage medium provided by the application has the same beneficial effects as the tobacco disease identification and prevention method provided by the above-mentioned embodiments, and will not be repeated here.

[0095] The application also provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of the tobacco disease identification and prevention method as described above.

[0096] The computer program product provided by the application can solve the technical problem of how to improve the accuracy of tobacco disease identification. Compared with the prior art, the computer program product provided by the application has the same beneficial effects as the tobacco disease identification and prevention method provided by the above-mentioned embodiments, and will not be repeated here.

[0097] The above-mentioned is only part of the embodiments of the application, and does not limit the patent scope of the application, and any equivalent structural transformation, direct / indirect application in other related technical fields within the technical concept of the application, and the contents of the specification and drawings are included in the patent protection scope of the application.

Claims

1. A method for identifying and controlling tobacco leaf diseases, characterized in that, The method includes: Construct and train a disease identification model based on physical perception constraints and hierarchical graph neural networks; Acquire an image of the tobacco leaf to be identified, process the image of the tobacco leaf to be identified using the disease identification model, and generate a disease category prediction result; Disease prevention and control plans are generated based on the preset multimodal large model and the disease category prediction results.

2. The method as described in claim 1, characterized in that, The steps for constructing and training a disease identification model based on physical perception constraints and hierarchical graph neural networks include: Acquire training image data of the target tobacco leaves; Disease features of the target tobacco leaf training image data are extracted based on physical perception constraints; Construct a hierarchical graph neural network architecture; A disease identification model is obtained by training the hybrid loss function of the hierarchical graph neural network architecture using the target tobacco leaf training image data and the disease features.

3. The method as described in claim 2, characterized in that, The steps for acquiring target tobacco leaf training image data include: Acquire raw tobacco leaf training image data; The original tobacco leaf training image data is adjusted according to the requirements of the preset network structure model to obtain the target tobacco leaf training image data.

4. The method as described in claim 2, characterized in that, The step of extracting disease features from the target tobacco leaf training image data based on physical perception constraints includes: The target tobacco leaf training image data is converted to a color space to extract abnormal color features of lesions; The edge texture of the lesion color abnormality features is enhanced using morphological gradient operators to obtain enhanced edge features; Establish a constraint equation for the area ratio of lesions to calculate the region range of the enhanced edge features; Disease features are generated based on the abnormal color characteristics of the lesions, the enhanced edge features, and the area range.

5. The method as described in claim 2, characterized in that, The step of training the hybrid loss function of the hierarchical graph neural network architecture using the target tobacco leaf training image data and the disease features includes: The predicted probability distribution results are obtained by using the target tobacco leaf training image data, the disease features, and the hierarchical graph neural network architecture. Obtain the true label distribution, calculate the difference between the predicted probability distribution and the true label distribution, and use the difference as the lesion distribution consistency loss; Adjust the boundary distance based on the disease characteristics to generate a boundary adjustment loss; A hybrid loss function is obtained by combining the lesion distribution consistency loss and the boundary adjustment loss.

6. The method as described in claim 1, characterized in that, The step of processing the tobacco leaf image to be identified using the disease identification model to generate a disease category prediction result includes: The image of the tobacco leaf to be identified is evenly divided into multiple blocks and defined as graph nodes, and the feature vectors of the graph nodes are extracted to obtain the initial features of the nodes; The connection weights between nodes are calculated based on the feature similarity and spatial position relationship of the initial features of the nodes, and a normalized adjacency matrix is ​​constructed based on the connection weights. The initial features of the nodes are processed by a graph convolutional layer with residual connections to obtain the low-level feature representation; A multi-head self-attention mechanism is used to capture the long-range dependencies represented by the low-level features and generate high-level features; The global information of the high-level features is aggregated to generate a graph-level representation, and the disease category prediction result is generated based on the graph-level representation.

7. The method as described in claim 1, characterized in that, The step of generating a disease prevention and control plan based on the preset multimodal large model and the disease category prediction results includes: The disease category prediction results are analyzed based on the preset multimodal large model and the preset expert knowledge base to obtain the analysis results; Based on the analysis results, a disease control plan is generated that includes recommended drugs, usage cycles, dosages, and control times.

8. A device for identifying and controlling tobacco leaf diseases, characterized in that, The device includes: The building module is used to build and train a disease identification model based on physical perception constraints and hierarchical graph neural networks; The prediction module is used to acquire images of tobacco leaves to be identified, process the images of tobacco leaves to be identified through the disease identification model, and generate disease category prediction results. The generation module is used to generate disease prevention and control schemes based on the preset multimodal large model.

9. A device for identifying and controlling tobacco leaf diseases, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the method for identifying and controlling tobacco diseases as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the method for identifying and controlling tobacco diseases as described in any one of claims 1 to 7.

Images