Artificial Intelligence-Based Plum Blossom Growth Cycle Disease Image Processing Method and System

Abstract

The present invention relates to a plum blossom growth cycle disease image processing method and system based on artificial intelligence, belonging to the technical field of plum blossom growth cycle disease image processing. The method includes the following steps: obtaining the original plum blossom disease image; extracting a multi-scale feature space through an image processing module; constructing a super-resolution enhancement structure with two modules in series, and respectively processing the features by two blur enhancement modules and then fusing and outputting; designing a reconstruction module including a convolutional layer, a blueprint separable convolution, and a residual connection, and generating high-resolution image features through feature fusion and upsampling; adopting global pooling and depth-shared convolution to construct an attention mechanism to capture disease semantic features; compressing the feature dimension through global average pooling, mapping it to a class space through a fully connected layer, and using a Softmax activation function to output classification probabilities. Through multi-level processing of feature enhancement-reconstruction-attention focusing, the present invention effectively improves the accuracy of the model in identifying plum blossom disease images.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing of diseases in the plum blossom growth cycle, and particularly relates to an image processing method and system for diseases in the plum blossom growth cycle based on artificial intelligence. Background Art

[0002] In the field of plum blossom cultivation, the identification and prevention of diseases in its growth cycle have always been an important direction in the research of forestry, agriculture and horticulture plants. In traditional production practices, the disease prevention and control of plum blossoms mainly rely on manual inspection to observe disease symptoms such as leaf discoloration and branch damage, and combine the experience of technicians to judge the type and severity of diseases. However, the number of plum blossom cultivation experts is small and their experience varies. There are problems such as strong subjectivity in plant disease identification and high misjudgment rate, often resulting in delays in plum blossom pest control or improper nutritional regulation in the growth cycle.

[0003] With the development of computer vision and deep learning technologies, image recognition methods based on convolutional neural networks have achieved remarkable results in the field of forestry and agricultural disease detection and identification. However, there are still some problems in the application of general image processing and recognition technologies in the identification of diseases in the plum blossom growth cycle: First, the phenotypic differences of the disease image features of plum blossom crops are relatively large in different growth stages, the disease images in the existing databases are relatively blurred, and the disease features are not obvious, resulting in low accuracy of the existing disease recognition methods; Second, there are complex natural condition backgrounds, light changes and leaf occlusion in the plum blossom planting scenarios, interfering with the model's recognition of plum blossom disease images; In addition, the model has poor generalization and recognition capabilities across different plum blossom varieties and climate regions, and there is a problem of low accuracy in the recognition of diseases in the plum blossom growth cycle.

[0004] Therefore, the present invention proposes an image processing method and system for diseases in the plum blossom growth cycle based on artificial intelligence to solve the problems of unclear disease images and complex backgrounds leading to low accuracy of disease model recognition in the existing plum blossom growth cycle disease prevention and control process. Summary of the Invention

[0005] In order to solve the above problems, the present invention provides an image processing method and system for diseases in the plum blossom growth cycle based on artificial intelligence.

[0006] To achieve the above object, the present invention is realized through the following technical solutions:

[0007] The present invention provides an image processing method for diseases in the plum blossom growth cycle based on artificial intelligence, including the following steps:

[0008] S1. Obtain plum blossom disease images ;

[0009] S2. The plum blossom disease images Input it into the plum blossom disease image processing module for processing to obtain the mapped feature space ;

[0010] S3. Construct a super-resolution enhancement module for plum blossom disease images, including two identical blurry image enhancement modules in series; input the mapped feature space into the first blurry image enhancement module for processing to obtain the first blurry image enhancement feature , and input the first blurry image enhancement feature into the second blurry image enhancement module to output the second blurry image enhancement feature ; splice and fuse the first blurry image enhancement feature and the second blurry image enhancement feature to output the super-resolution enhancement feature of the plum blossom disease image ;

[0011] S4. Construct a reconstruction module for plum blossom disease images, including a convolutional layer, a blueprint separable convolutional layer, a global residual connection layer, and an upsampling layer; the super-resolution enhancement feature of the plum blossom disease image is processed through the convolutional layer and the blueprint separable convolutional layer to obtain an intermediate feature , and the intermediate feature and the mapped feature space are fused through the global residual connection layer to retain the original image information, obtaining a fused global feature , and the fused global feature reconstructs the plum blossom disease image feature through the upsampling layer to obtain the super-resolution plum blossom disease image feature ;

[0012] S5. Construct a super-resolution image attention semantic feature capture module, including a global pooling layer, a depth-shared convolutional layer, a feature splicing operation, and an activation function; the super-resolution plum blossom disease image feature is processed through the super-resolution image attention semantic feature capture module to obtain the super-resolution image attention semantic feature ;

[0013] S6. Construct a plum blossom disease image recognition output module, input the super-resolution image attention semantic feature into this module, first perform global average pooling on the spatial dimension of the feature map through the global pooling layer to output the global average pooling feature , map the feature vector to the category space through the fully connected layer for the global average pooling feature to output the mapped feature , and finally input the mapped feature into The activation function is used to obtain the probability distribution of each sample and the final recognition and classification prediction result is obtained .

[0014] Furthermore, step S2 specifically includes:

[0015] S21. The low-resolution plum blossom disease image After performing the low-resolution image channel replication operation and replicating times along the channel dimension, the replicated low-resolution image is obtained by splicing along the channel dimension , and the formula is expressed as follows:

[0016] ,

[0017] where represents the splicing operation, represents the number of disease samples input into the model, represents the height of the disease image, represents the width of the disease image, represents the number of channels of the disease image, ;

[0018] S22. The replicated low-resolution image Undergoes a low-resolution image feature mapping operation. The replicated low-resolution image is mapped using a blueprint separable convolution to obtain a mapped feature space , and the blueprint separable convolution includes a convolutional layer with a kernel size of and a depthwise separable convolutional layer with a kernel size of , and the formula is expressed as follows:

[0019] ,

[0020] where represents the blueprint separable convolution, represents the convolutional layer with a kernel size of , represents the depthwise separable convolutional layer with a kernel size of .

[0021] Furthermore, step S3 specifically includes:

[0022] S31. The mapped feature space Passes through a convolutional layer with a kernel size of of the first blurred image enhancement module, and is divided into two sub-features with the same number of channels according to the channels of the mapped feature space , the first sub-feature and the second sub-feature ; The second sub - feature Pass through a convolutional layer with a kernel size of and split it into a third sub - feature and a fourth sub - feature with the same number of channels along the channels ; The third sub - feature Pass through a convolutional layer with a kernel size of to obtain the first spatial feature . Concatenate and fuse the first spatial feature with the fourth sub - feature , and use a convolutional layer with a kernel size of to fuse the features, obtaining the first output feature . The formula is as follows:

[0023] ,

[0024] ,

[0025] ,

[0026] ,

[0027] where represents the operation of splitting the feature according to the channel feature, represents a depth - separable convolutional layer with a kernel size of , represents the activation function operation;

[0028] S32. Pass the first output feature through a convolutional layer with a kernel size of , and split it into a fifth sub - feature and a sixth sub - feature along the channels of the first output feature . The sixth sub - feature passes through a convolutional layer with a kernel size of and is split into a seventh sub - feature and an eighth sub - feature with the same number of channels along the channels; Use a convolutional layer with a kernel size of to perform convolutional operations layer by layer on the seventh sub - featureto obtain the second spatial feature . Concatenate and fuse the second spatial feature with the eighth sub - feature , and use a convolutional layer with a kernel size of to fuse the features, obtaining the second output feature . The formula is as follows:

[0029] ,

[0030] ,

[0031] ,

[0032] ,

[0033] Among them, represents a depthwise separable convolutional layer with a convolutional kernel size of ;

[0034] S33. Input the second output feature into the blueprint separable convolution for processing to obtain a third output feature , and the formula is expressed as follows:

[0035] ;

[0036] The third output feature , the first sub-feature and the fifth sub-feature are concatenated and fused and then adjusted by a convolutional layer with a convolutional kernel size of to obtain a first blurred image enhancement feature , and the formula is expressed as follows:

[0037] ;

[0038] Input the first blurred image enhancement feature into the second blurred image enhancement module to obtain a second blurred image enhancement feature , concatenate and fuse the features and to obtain a plum blossom disease image super-resolution enhancement feature .

[0039] Furthermore, step S4 specifically includes:

[0040] S41. The plum blossom disease image super-resolution enhancement feature passes through a convolutional layer with a convolutional kernel size of and an activation function operation to obtain an initial reconstruction feature , and input the initial reconstruction feature into the blueprint separable convolutional layer for processing to obtain an intermediate feature , and the formula is expressed as follows:

[0041] ,

[0042] ,

[0043] Among them, represents the blueprint separable convolution, represents the activation function operation;

[0044] S42. Intermediate features and the mapped feature space go through the global residual connection operation to obtain the fused global features , and the fused global features go through the upsampling operation to reconstruct the plum blossom disease image features and obtain the super-resolution plum blossom disease image features , and the formula is as follows:

[0045] ,

[0046] ,

[0047] Among them, represents the element-wise addition operation, represents the operation of reconstructing the super-resolution image.

[0048] Furthermore, step S5 specifically includes:

[0049] S51. The super-resolution plum blossom disease image features go through the global average pooling layer to perform global pooling along the height and width dimensions respectively, obtaining the sub-features decomposed along the height and the sub-features decomposed along the width , and the sub-features and are evenly divided into groups along the channel dimension, obtaining the sub-feature groups decomposed along the height and the sub-feature groups decomposed along the width , and the formula is as follows:

[0050] ,

[0051] ,

[0052] ,

[0053] ,

[0054] Among them, represents the number of channels of the sub-features, represents the width of the sub-features decomposed along the height, represents the height of the sub-features decomposed along the width, Represents a global pooling operation, Indicates decomposing the features by compressing them along the height dimension, Indicates decomposing the features by compressing them along the width dimension, And Respectively represent slicing the sub - features And Grouping and slicing the features according to the feature channel dimension, ; When At this time, The channel dimension index of is When , The channel dimension index of is And so on, The operation is the same;

[0055] S52. The first grouped features decomposed along the height , the second grouped features , the third grouped features And the fourth grouped features Respectively pass through a depth - separable 1D convolutional layer with a kernel size of To obtain the processed first grouped features decomposed along the height , the second grouped features , the third grouped features And the fourth grouped features , The formula is as follows:

[0056] ,

[0057] ,

[0058] ,

[0059] ,

[0060] Among them, 、 、 Respectively represent depth - separable convolution operations with kernel sizes of 、 、 ; The processing process of the sub - feature grouping Decomposed along the width is the same, and the processed first grouped features decomposed along the width , the second grouped features , the third grouped features And the fourth grouped features Are obtained;

[0061] The first grouped feature decomposed along the height after processing , the second grouped feature , the third grouped feature and the fourth grouped feature are processed through splicing fusion and using group normalization to obtain the first new feature ; Similarly, the first grouped feature decomposed along the width after processing , the second grouped feature , the third grouped feature and the fourth grouped feature undergo the same operation to obtain the second new feature ; The first new feature and the second new feature are input into the activation function to obtain the first spatial attention feature map and the second spatial attention feature map . The first spatial attention feature map , the second spatial attention feature map and the super-resolution plum blossom disease image feature are fused to obtain the super-resolution image attention semantic feature , and the formula is as follows:

[0062] ,

[0063] ,

[0064] ,

[0065] ,

[0066] ,

[0067] Among them, represents the group normalization operation, represents the Sigmoid activation function, represents the element-wise multiplication operation.

[0068] Furthermore, step S6 specifically includes:

[0069] The super-resolution image attention semantic feature undergoes global average pooling on the spatial dimension of the feature map through the global pooling layer, and outputs the global average pooling feature . The global average pooling feature is processed through the fully connected layer to obtain the output mapping feature . The mapping feature is input into The activation function is used to obtain the probability distribution of each sample and obtain the final recognition and classification prediction result The formula is expressed as follows:

[0070] ,

[0071] ,

[0072] ,

[0073] ,

[0074] Among them, represents the global average pooling operation represents the fully connected layer represents using the maximum index value as the class label of the image finally recognized by the model represents traversing the maximum index value in the first dimension

[0075] The present invention also provides an artificial intelligence-based plum blossom growth cycle disease image processing system, which executes the artificial intelligence-based plum blossom growth cycle disease image processing method, including:

[0076] Data acquisition module: used to acquire plum blossom disease images;

[0077] Plum blossom disease image processing module: used to process the plum blossom disease images to obtain the mapped feature space;

[0078] Plum blossom disease image super-resolution enhancement module: used to perform super-resolution enhancement on the mapped feature space to obtain the plum blossom disease image super-resolution enhancement features;

[0079] Plum blossom disease image reconstruction module: used to reconstruct the plum blossom disease image super-resolution enhancement features to obtain the super-resolution plum blossom disease image features;

[0080] Super-resolution image attention semantic feature capture module: used to capture the semantic features in the super-resolution plum blossom disease image features to obtain the super-resolution image attention semantic features;

[0081] Plum blossom disease image recognition and output module: used to input the super-resolution image attention semantic features into this module to obtain the probability distribution of each sample and obtain the final recognition and classification prediction result

[0082] The advantages of the present invention are:

[0083] The present invention uses a plum blossom disease image super-resolution enhancement module to perform super-resolution processing on disease images, and uses a plum blossom disease image reconstruction module to convert low-resolution images that are blurred and difficult to identify into high-resolution images that are easy to identify, which helps the model to identify and classify plum blossom disease images, improves the recognition accuracy of the model. Through the super-resolution image attention semantic feature capture module, local and global spatial semantic features of disease images are captured, and the low-level texture information of disease images is deeply mined, enhancing the model's ability to process disease images in complex scenarios, thereby effectively improving the accuracy of the model in identifying plum blossom disease images. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The 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 to the present invention.

[0085] Figure 1 is a flowchart of the steps of the method of the present invention;

[0086] Figure 2 is a comparison before and after processing of plum blossom disease images in the method of the present invention;

[0087] Figure 3 is an effect diagram of a comparative experiment of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0088] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0089] Embodiment 1

[0090] In this embodiment, as Figure 1 shown, the present invention provides an artificial intelligence-based plum blossom growth cycle disease image processing method, and the specific steps include:

[0091] S1. Obtain plum blossom disease images ;

[0092] S2. Input the plum blossom disease images into the plum blossom disease image processing module for processing to obtain a mapped feature space ;

[0093] Specifically, S21. The low-resolution plum blossom disease images are subjected to a low-resolution image channel replication operation, and replicated in the channel dimension After that, concatenate along the channel dimension to obtain the replicated low-resolution image. , which is expressed by the formula as follows:

[0094] ,

[0095] Among them, represents the concatenation operation, represents the number of disease samples input to the model, represents the height of the disease image, represents the width of the disease image, represents the number of channels of the disease image, ;

[0096] S22. Perform a low-resolution image feature mapping operation on the replicated low-resolution image , and use a blueprint separable convolution to map the replicated low-resolution image to obtain a mapped feature space , and the blueprint separable convolution includes a convolutional layer with a kernel size of and a depthwise separable convolutional layer with a kernel size of , which is expressed by the formula as follows:

[0097] ,

[0098] Among them, represents the blueprint separable convolution, represents the convolutional layer with a kernel size of , represents the depthwise separable convolutional layer with a kernel size of .

[0099] S3. Construct a plum blossom disease image super-resolution enhancement module, which includes two cascaded blurred image enhancement modules with the same structure; input the mapped feature space into the first blurred image enhancement module for processing to obtain the first blurred image enhancement feature , input the first blurred image enhancement feature into the second blurred image enhancement module, and output the second blurred image enhancement feature ; concatenate and fuse the first blurred image enhancement feature and the second blurred image enhancement feature to output the plum blossom disease image super-resolution enhancement feature ;

[0100] Specifically, S31. The mapped feature space passes through a convolutional layer with a kernel size of in the first blurred image enhancement module, according to the mapped feature space The channels are split into two sub - features with the same number of channels. The first sub - feature and the second sub - feature ; The second sub - feature passes through a convolutional layer with a kernel size of and is split along the channels into a third sub - feature and a fourth sub - feature with the same number of channels; The third sub - feature is processed through a convolutional layer with a kernel size of to obtain a first spatial feature . The first spatial feature is concatenated and fused with the fourth sub - feature , and a convolutional layer with a kernel size of is used to fuse the features, obtaining a first output feature . The formula is as follows:

[0101] ,

[0102] ,

[0103] ,

[0104] ,

[0105] where, represents the operation of splitting the feature according to the channel feature, represents a depth - separable convolutional layer with a kernel size of , represents the activation function operation;

[0106] S32. The first output feature passes through a convolutional layer with a kernel size of and is split into a fifth sub - feature and a sixth sub - feature along the channels of the first output feature . The sixth sub - feature passes through a convolutional layer with a kernel size of and is split along the channels into a seventh sub - feature and an eighth sub - feature with the same number of channels; The seventh sub - feature is convolved using a convolutional layer with a kernel size of layer by layer to obtain a second spatial feature . The second spatial feature is concatenated and fused with the eighth sub - feature , and a convolutional layer with a kernel size of The convolutional layer fuses the features to obtain the second output feature , which is expressed by the following formula:

[0107] ,

[0108] ,

[0109] ,

[0110] ,

[0111] where represents a depthwise separable convolutional layer with a convolutional kernel size of ;

[0112] S33. Input the second output feature into the blueprint separable convolution for processing to obtain the third output feature , which is expressed by the following formula:

[0113] ;

[0114] The third output feature , the first sub - feature and the fifth sub - feature are concatenated and fused and then adjusted by a convolutional layer with a convolutional kernel size of to obtain the first blurred image enhancement feature , which is expressed by the following formula:

[0115] ;

[0116] Input the first blurred image enhancement feature into the second blurred image enhancement module to obtain the second blurred image enhancement feature , and concatenate and fuse the features and to obtain the plum blossom disease image super - resolution enhancement feature .

[0117] S4. Construct a plum blossom disease image reconstruction module, including a convolutional layer, a blueprint separable convolutional layer, a global residual connection layer, and an upsampling layer; the plum blossom disease image super - resolution enhancement feature is processed by a convolutional layer and a blueprint separable convolutional layer to obtain an intermediate feature , and the intermediate feature and the mapped feature space are fused through the global residual connection layer to retain the original image information, obtaining the fused global feature , and the fused global feature Reconstruct the plum blossom disease image features through the upsampling layer to obtain the super-resolution plum blossom disease image features ;

[0118] Specifically, S41. The super-resolution enhanced features of the plum blossom disease image Pass through the convolutional layer with a convolutional kernel size of and the activation function operation to obtain the initial reconstructed features , and input the initial reconstructed features into the blueprint separable convolutional layer for processing to obtain the intermediate features , and the formula is as follows:

[0119] ,

[0120] ,

[0121] Among them, represents the blueprint separable convolution, represents the activation function operation;

[0122] S42. The intermediate features and the mapped feature space pass through the global residual connection operation to obtain the fused global features , and the fused global features pass through the upsampling operation to reconstruct the plum blossom disease image features to obtain the super-resolution plum blossom disease image features , and the formula is as follows:

[0123] ,

[0124] ,

[0125] Among them, represents the element-wise addition operation, represents the operation of reconstructing the super-resolution image.

[0126] S5. Construct a super-resolution image attention semantic feature capture module, including a global pooling layer, a depth-shared convolutional layer, a feature concatenation operation, and an activation function; the super-resolution plum blossom disease image features are processed through the super-resolution image attention semantic feature capture module to obtain the super-resolution image attention semantic features ;

[0127] Specifically, S51. The super-resolution plum blossom disease image features After global average pooling is performed along the height and width dimensions respectively, sub-features decomposed along the height are obtained and sub-features decomposed along the width . The sub-features and are evenly divided into groups along the channel dimension, obtaining grouped sub-features decomposed along the height and grouped sub-features decomposed along the width . The formula is as follows:

[0128] ,

[0129] ,

[0130] ,

[0131] ,

[0132] where represents the number of channels of the sub-feature, represents the width of the sub-feature decomposed along the height, represents the height of the sub-feature decomposed along the width, represents the global pooling operation, represents compressing and decomposing the feature along the height dimension, represents compressing and decomposing the feature along the width dimension, and respectively represent slicing and grouping the sub-features and along the feature channel dimension. When , the channel dimension index of is , when , the channel dimension index of , is , and so on. The operation of is the same;

[0133] S52. The first grouped feature decomposed along the height , the second grouped feature , the third grouped feature and the fourth grouped feature are respectively passed through a depthwise separable 1D convolutional layer with a kernel size of , obtaining the processed first grouped feature decomposed along the height , the second grouped feature , the third grouped feature and the fourth grouped feature , which is expressed by the formula as follows:

[0134] ,

[0135] ,

[0136] ,

[0137] ,

[0138] Among them, 、 、 respectively represent the depthwise separable convolution operations with convolution kernel sizes of 、 、 ; The processing process of the sub - feature groups decomposed along the width is the same, and the first grouped feature decomposed along the width after processing, the second grouped feature decomposed along the width after processing, the third grouped feature decomposed along the width after processing, and the fourth grouped feature ;

[0139] S53. The first grouped feature decomposed along the height after processing, the second grouped feature decomposed along the height after processing, the third grouped feature decomposed along the height after processing, and the fourth grouped feature are processed through splicing fusion and grouped normalization to obtain the first new feature ; Similarly, the first grouped feature decomposed along the width after processing, the second grouped feature decomposed along the width after processing, the third grouped feature decomposed along the width after processing, and the fourth grouped feature undergo the same operation to obtain the second new feature ; The first new feature and the second new feature are input into the activation function to obtain the first spatial attention feature map and the second spatial attention feature map . The first spatial attention feature map , the second spatial attention feature map and the super - resolution plum blossom disease image feature are fused to obtain the super - resolution image attention semantic feature , which is expressed by the formula as follows:

[0140] ,

[0141] ,

[0142] ,

[0143] ,

[0144] ,

[0145] Among them, represents the group normalization operation, represents the Sigmoid activation function, represents the element-wise multiplication operation.

[0146] S6. Construct a plum blossom disease image recognition output module, and input the super-resolution image attention semantic feature into this module. First, perform global average pooling on the spatial dimension of the feature map through the global pooling layer to output the global average pooling feature . Map the global average pooling feature to the class space through the fully connected layer to output the mapped feature . Finally, input the mapped feature into the activation function to obtain the probability distribution of each sample , and obtain the final recognition classification prediction result .

[0147] Specifically, the super-resolution image attention semantic feature performs global average pooling on the spatial dimension of the feature map through the global pooling layer to output the global average pooling feature . The global average pooling feature is processed through the fully connected layer to obtain the output mapped feature . Input the mapped feature into the activation function to obtain the probability distribution of each sample , and obtain the final recognition classification prediction result , and the formula is expressed as follows:

[0148] ,

[0149] ,

[0150] ,

[0151] ,

[0152] Among them, represents the global average pooling operation, Represents a fully connected layer, Represents the maximum index value as the class label of the image finally recognized by the model, Represents traversing the maximum index value in the first dimension.

[0153] Example 2

[0154] To verify the effectiveness of the proposed artificial intelligence-based plum blossom growth cycle disease recognition method and system in the direction of disease image processing and recognition, the method was experimentally compared with existing forestry and agricultural disease image processing technologies under the same conditions and environment. The obtained experimental results were analyzed and demonstrated in detail to verify the effectiveness of the proposed method in the direction of plum blossom growth cycle disease image processing.

[0155] In the comparative experiment, three currently mainstream image processing models were selected as the comparative models to prove the performance effect of the proposed method. The three comparative models used in the experiment were: ConvNeXt is a hybrid model improved based on the CNN architecture. By combining the local feature extraction ability of CNN and the global modeling ability of Transformer, it achieves high accuracy in image classification tasks, but its generalization ability in small sample tasks and complex scenarios is poor; EfficientViT adopts a multi-scale local attention mechanism and depthwise separable convolution, and can achieve classification accuracy comparable to that of large models, but its classification ability for long-tailed distribution data is weak and it depends on pre-trained weights; SwinTransformer enhances the multi-scale feature expression ability while reducing the computational complexity through sliding window self-attention, but the window size and hierarchical structure need to be finely adjusted during the training process, and the model is relatively complex.

[0156] In the comparative experiment of the proposed artificial intelligence-based plum blossom growth cycle disease image processing method, four experimental indicators were used to evaluate the effectiveness of the proposed method: Accuracy (Acc) reflects the overall classification accuracy of the proposed model for plum blossom healthy and disease samples; Precision (Pre) measures the proportion of samples predicted as plum blossom diseases that are actually diseases; Recall represents the proportion of truly plum blossom disease samples that are correctly recognized; F1-score is suitable for evaluating scenarios with uneven distribution of plum blossom disease data by taking the harmonic mean of precision and recall.

[0157] To this end, an image dataset for identifying diseases in the plum blossom growth cycle was established in the comparative experiment. This dataset collected images through web crawlers from public agricultural websites, plant pathology forums, and plum blossom cultivation-related platforms, covering the healthy and diseased states of plum blossoms in various complex scenarios, and recording the performance characteristics at different growth stages. The dataset contains 13 categories (including healthy states and disease types), and the number of images in each category shows a long-tailed distribution. In the experiment, the images of each category in the dataset were divided into a training set and a test set according to a ratio of 8:2 to ensure the generalization ability and evaluation reliability of the model in diverse scenarios.

[0158] Table 1 Comparative results of the proposed method in the dataset

[0159]

[0160] The proposed method was experimentally verified in the plum blossom growth cycle disease dataset. Figure 2 High-resolution images of plum blossom diseases after being processed by the proposed method are shown. The specific experimental results are as shown in Table 1 and Figure 3 as follows. The experimental results show that the proposed method performs optimally in the image processing and recognition tasks of plum blossom growth cycle diseases. Through this method, relatively blurred plum blossom disease images can be processed into high-resolution disease images, which helps the model accurately identify the characteristics of plum blossom disease images, and all four experimental indicators are significantly higher than those of other models. In the Figure 3 comparative test effect diagram, the recognition accuracies of the proposed method for chlorosis, perforation disease, and powdery mildew in plum blossom diseases reach 86%, 87%, and 82% respectively, all of which are the best results among the comparative methods. Based on the above two experimental results, the effectiveness of the proposed method in the field of plum blossom growth cycle disease image processing and recognition is proved.

[0161] Example 3

[0162] This example provides an artificial intelligence-based plum blossom growth cycle disease image processing system that executes the artificial intelligence-based plum blossom growth cycle disease image processing method described in Example 1, including:

[0163] Data acquisition module: used to acquire plum blossom disease images;

[0164] Plum blossom disease image processing module: used to process plum blossom disease images to obtain a mapped feature space;

[0165] Plum blossom disease image super-resolution enhancement module: used to perform super-resolution enhancement on the mapped feature space to obtain plum blossom disease image super-resolution enhancement features;

[0166] Plum blossom disease image reconstruction module: used to reconstruct the plum blossom disease image super-resolution enhancement features to obtain super-resolution plum blossom disease image features;

[0167] Super-resolution image attention semantic feature capture module: used to capture the semantic features in the features of the super-resolution plum disease image, and obtain the super-resolution image attention semantic features;

[0168] Plum disease image recognition output module: used to input the super-resolution image attention semantic features into this module, obtain the probability distribution of each sample, and obtain the final recognition and classification prediction result.

[0169] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An artificial intelligence-based plum blossom growth cycle disease image processing method, characterized in that: The following steps are involved: S1. Obtaining plum blossom disease images ; S2. Plum blossom disease image Input to the plum blossom disease image processing module for processing to obtain the mapped feature space ; S3. Construct a super-resolution enhancement module for plum blossom disease images, including two fuzzy image enhancement modules with the same structure in series; map the feature space Input to the first fuzzy image enhancement module for processing to obtain the first fuzzy image enhancement feature , the first blurred image enhancement feature Input to the second fuzzy image enhancement module, output the second fuzzy image enhancement feature ; The first blurred image is enhanced and the second fuzzy image enhancement feature Splicing and fusion to output super-resolution enhancement features of plum blossom disease images ; The specific steps are as follows: S31. Mapped feature space The convolution kernel size after the first blurred image enhancement module is The convolution layer, according to the mapped feature space The channels are divided into two sub-features with the same number of channels. The first sub-feature and the second sub-feature ; Second sub-feature After convolution again, the kernel size is The convolutional layer is divided into third sub-features with the same number of channels along the channel and the fourth sub-characteristic ; The third sub-feature After convolution, the kernel size is The convolution layer is processed to obtain the first spatial feature , the first spatial feature With the fourth sub-characteristic Perform splicing and fusion, and use the convolution kernel size of The convolutional layer fusion features are used to obtain the first output feature , the formula is as follows: ， ， ， ， in, Indicates that the features are split according to channel features. Indicates that the convolution kernel size is Depthwise separable convolutional layers, express Activation function operation; S32. The first output feature After convolution, the kernel size is The convolution layer, according to the first output feature The channel splits it into the fifth sub-feature and the sixth sub-characteristic , the sixth sub-feature After convolution, the kernel size is The convolutional layer is divided into the seventh sub-feature with the same number of channels along the channel and the eighth sub-characteristic ; The seventh sub-feature The convolution kernel size is used Convolution operation is performed layer by layer to obtain the second spatial feature , the second spatial feature With the eighth sub-characteristic Perform splicing and fusion, and use the convolution kernel size of The convolutional layer fusion features are used to obtain the second output feature , the formula is as follows: ， ， ， ， in, Indicates that the convolution kernel size is Depthwise separable convolutional layers; S33. The second output feature Input to the blueprint separable convolution for processing to obtain the third output feature , the formula is as follows: ； The third output characteristic , the first sub-feature And the fifth sub-characteristic After splicing and fusion and convolution kernel size is The convolution layer adjusts the features to obtain the first blurred image enhancement feature , the formula is as follows: ； The first blurred image is enhanced Input to the second fuzzy image enhancement module to obtain the second fuzzy image enhancement feature , the feature and Splicing and fusion to obtain super-resolution enhancement features of plum blossom disease images ; S4. Construct a plum blossom disease image reconstruction module, including convolutional layer, blueprint separable convolutional layer, global residual connection layer and upsampling layer; super-resolution enhancement feature of plum blossom disease image After processing by the convolution layer and the blueprint separable convolution layer, the intermediate features are obtained , intermediate features The feature space with the mapping After the global residual connection layer fusion, the original image information is retained to obtain the fused global features , integrating global features The plum blossom disease image features are reconstructed through the upsampling layer to obtain the super-resolution plum blossom disease image features. ; The specific steps are as follows: S41. Super-resolution enhancement features of the plum blossom disease image After convolution, the kernel size is The convolutional layers and Activation function operation to obtain the initial reconstruction features , the initial reconstruction feature Input to the blueprint separable convolutional layer for processing to obtain intermediate features , the formula is as follows: ， ， in, represents the blueprint separable convolution, express Activation function operation; S42. Intermediate features The feature space with the mapping After the global residual connection operation, the fused global features are obtained , integrating global features After upsampling, the image features of plum blossom diseases are reconstructed to obtain super-resolution image features of plum blossom diseases. , the formula is as follows: ， ， in, represents the element-by-element addition operation, Represents the operation of reconstructing super-resolution images; S5. Construct a super-resolution image attention semantic feature capture module, including a global pooling layer, a deep shared convolutional layer, a feature splicing operation, and Activation function; Super-resolution plum blossom disease image features After being processed by the super-resolution image attention semantic feature capture module, the super-resolution image attention semantic feature is obtained ; S6. Construct a plum blossom disease image recognition output module to focus on the semantic features of super-resolution images The input to this module first passes through the global pooling layer to perform global average pooling on the spatial dimension of the feature map, and outputs the global average pooling feature. , the global average pooling feature Map the feature vector to the category space through the fully connected layer and output the mapping feature , and finally the mapping feature Input to Activation function to obtain the probability distribution of each sample , and obtain the final recognition and classification prediction results .

2. The method for processing plum blossom growth cycle diseases image based on artificial intelligence according to claim 1 is characterized in that: Step S2 specifically includes: S21. Low-resolution image of plum blossom disease After the low-resolution image channel copy operation, copy in the channel dimension After that, the copied low-resolution image is spliced ​​along the channel dimension. , the formula is as follows: ， in, Represents a splicing operation, represents the number of disease samples input into the model, Indicates the height of the disease image, represents the width of the diseased image, Indicates the number of channels of the disease image, ; S22. The copied low-resolution image Perform low-resolution image feature mapping operations, use blueprint separable convolution to map the copied low-resolution image, and obtain the mapped feature space , the blueprint separable convolution includes a convolution kernel size of The convolution layer and convolution kernel size are The depth-separable convolutional layer is expressed as follows: ， in, represents the blueprint separable convolution, Indicates that the convolution kernel size is The convolutional layer, Indicates that the convolution kernel size is The depth of the separable convolutional layer.

3. The method for processing plum blossom growth cycle diseases image based on artificial intelligence according to claim 2 is characterized in that: Step S5 specifically includes: S51. Super-resolution plum blossom disease image features After the global average pooling layer, global pooling is performed along the height and width dimensions to obtain sub-features decomposed along the height. and sub-features decomposed along the width , the sub-feature and Divided equally along the channel dimension Group, get the sub-feature grouping along the height decomposition and sub-feature grouping along width decomposition , the formula is as follows: ， ， ， ， in, represents the number of channels of the sub-features, represents the width of the sub-feature decomposed along the height, represents the height of the sub-features decomposed along the width, represents the global pooling operation, Indicates that the features are compressed and decomposed according to the height dimension. Indicates that the features are compressed and decomposed according to the width dimension. and Respectively represent the characteristics of the pair and Group and slice the features according to the feature channel dimension. ;when hour, The channel dimension index is ,when , The channel dimension index is , and so on, The operation is the same; S52. Decompose the first grouping feature along the height , the second grouping feature , the third grouping feature And the fourth grouping feature After the convolution kernel size is The depth of the separable 1D convolution layer obtains the first grouping feature after processing along the height decomposition , the second grouping feature , the third grouping feature And the fourth grouping feature , the formula is as follows: ， ， ， ， in, , , They represent the convolution kernel sizes respectively. , , Depthwise separable convolution operation; sub-feature grouping along width decomposition The processing process is similar to that of the first grouping feature decomposed along the width. , the second grouping feature , the third grouping feature And the fourth grouping feature ; S53. The first grouping feature after processing along the height decomposition , the second grouping feature , the third grouping feature And the fourth grouping feature The first new feature is obtained by concatenating and fusion and using group normalization. ; Similarly, the first grouping feature after processing along the width decomposition , the second grouping feature , the third grouping feature And the fourth grouping feature After the same operation, the second new feature is obtained ; The first new feature And the second new feature Input to Activation function, respectively get the first spatial attention feature map and the second spatial attention feature map , the first spatial attention feature map , Second spatial attention feature map And super-resolution plum blossom disease image features Fusion is performed to obtain the super-resolution image attention semantic features , the formula is as follows: ， ， ， ， ， in, represents the group normalization operation, represents the Sigmoid activation function, Represents an element-wise multiplication operation.

4. The method for processing plum blossom growth cycle diseases image based on artificial intelligence according to claim 3 is characterized in that: Step S6 specifically includes: Super-resolution image attention semantic features The spatial dimension of the feature map is globally averaged pooled through the global pooling layer, and the global average pooling feature is output , global average pooling feature After processing by the fully connected layer, the output mapping features are obtained , the mapping feature Input to Activation function to obtain the probability distribution of each sample , and obtain the final recognition and classification prediction results , the formula is as follows: ， ， ， ， in, represents the global average pooling operation, represents the fully connected layer, Indicates that the maximum index value is used as the category label of the model's final recognition image. Indicates traversing the maximum index value in the first dimension.

5. An artificial intelligence-based plum blossom growth cycle disease image processing system, which executes the artificial intelligence-based plum blossom growth cycle disease image processing method as claimed in claim 1, characterized in that: include: Data acquisition module: used to obtain plum blossom disease images; Plum blossom disease image processing module: used to process plum blossom disease images and obtain mapped feature space; Plum blossom disease image super-resolution enhancement module: used to perform super-resolution enhancement on the mapped feature space to obtain super-resolution enhancement features of plum blossom disease images; Plum blossom disease image reconstruction module: used to reconstruct the super-resolution enhancement features of the plum blossom disease image to obtain the super-resolution plum blossom disease image features; Super-resolution image attention semantic feature capture module: used to capture the semantic features in the super-resolution plum blossom disease image features to obtain super-resolution image attention semantic features; Plum blossom disease image recognition output module: used to input the super-resolution image attention semantic features into this module, obtain the probability distribution of each sample, and obtain the final recognition and classification prediction results.

Images