Dried orange peel year identification method, system and equipment and medium

Abstract

The invention discloses a dried orange peel year identification method, system and device and a medium, and the method comprises the steps: firstly inputting a dried orange peel image marked with a year label into an initial year identification model, and then carrying out the feature extraction of the input dried orange peel image, and obtaining initial feature mapping; and on this basis, multi-view features in the initial feature mapping are further extracted, associated features among the features are identified, and year prediction is carried out by using the associated features as a basis to obtain a year prediction result. Further, the prediction result is compared with a year label, parameters of the initial year recognition model are adjusted accordingly, and finally an optimized year recognition model is obtained. In practical application, an accurate year prediction result can be obtained only by inputting a to-be-identified pericarpium citri reticulatae image into the trained year identification model. According to the embodiment of the invention, the multi-view features of the pericarpium citri reticulatae are comprehensively captured, and the associated features of the features are deeply analyzed, so that the accuracy and efficiency of year recognition are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to but are not limited to the technical field of tangerine peel year identification, and in particular, to a tangerine peel year identification method, system, device and medium. Background Art

[0002] In the traditional practice of identifying the year of tangerine peel, people mainly rely on manual experience and sensory evaluation to determine the year of tangerine peel, which is easily interfered by human factors, thus causing inaccuracy and inconsistency in the results. Especially when facing large-scale testing needs, it is not only time-consuming and laborious, but also requires comprehensive consideration of the front and back features of tangerine peel. The testing process is cumbersome and complicated, making it difficult for tangerine peel year identification to be flexible, efficient and accurate. As an agricultural product, the characteristics of tangerine peel are deeply affected by natural environmental conditions. Directly applying the general network model to the tangerine peel epidermis for feature extraction and classification often fails to accurately capture the subtle differences in tangerine peel that are unique to different years, and it is difficult to ensure the accuracy and reliability of the year identification results. Summary of the invention

[0003] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0004] The embodiment of the present application provides a method for identifying the year of tangerine peel, which can more comprehensively capture the multi-view features of tangerine peel, thereby improving the accuracy and efficiency of tangerine peel year identification.

[0005] In a first aspect, an embodiment of the present application provides a method for identifying the year of tangerine peel, comprising: obtaining a tangerine peel image, inputting the tangerine peel image into an initial year recognition model, wherein the tangerine peel image is provided with a year label; performing feature extraction on the tangerine peel image to obtain an initial feature map; extracting multi-view features of the initial feature map, and identifying correlation features between different multi-view features; performing year prediction based on the correlation features to obtain a year prediction result; adjusting the parameters of the initial year recognition model based on the year prediction result and the year label to obtain a year recognition model; inputting the tangerine peel image to be identified into the year recognition model to perform year prediction to obtain a year prediction result for the tangerine peel image to be identified.

[0006] In combination with the first aspect, in an embodiment of the present application, the multi-view feature extraction of the initial feature map includes: extracting local features of the initial feature map, normalizing and feature mapping the local features, and obtaining a first feature map; performing channel attention calculation on the first feature map to obtain a second feature map; performing spatial attention calculation on the second feature map to obtain a third feature map; and aggregating the initial feature map and the third feature map to obtain multi-view features.

[0007] In combination with the first aspect, in an embodiment of the present application, the identifying of correlation features between different multi-view features includes: performing feature compression processing on the multi-view features to obtain integrated features; performing average pooling, nonlinear combination and feature mapping processing on the integrated features to obtain a fourth feature map; and performing feature integration on the initial feature map and the fourth feature map to obtain correlation features.

[0008] In combination with the first aspect, in an embodiment of the present application, the multi-view features include front view features and back view features of the tangerine peel image; the feature compression processing of the multi-view features to obtain integrated features includes: compressing the front view features and back view features of the tangerine peel image into a single feature to obtain an integrated feature.

[0009] In combination with the first aspect, in an embodiment of the present application, before the tangerine peel image is input into the initial year recognition model, the steps include: denoising the tangerine peel image; and unifying the image size of the tangerine peel image.

[0010] In a second aspect, an embodiment of the present application provides a tangerine peel year recognition system, comprising: an image acquisition module, used to obtain a tangerine peel image; a tangerine peel year recognition module, used to receive the tangerine peel image, and perform the following tangerine peel year recognition method on the tangerine peel image:

[0011] Inputting the tangerine peel image into an initial year recognition model, wherein the tangerine peel image is provided with a year label;

[0012] Extracting features from the tangerine peel image to obtain an initial feature map;

[0013] Extracting multi-view features of the initial feature map, identifying associated features between different multi-view features, the multi-view features including front view features and back view features of the tangerine peel image;

[0014] Perform year prediction according to the correlation characteristics to obtain year prediction results;

[0015] Adjusting the parameters of the initial year recognition model according to the year prediction result and the year label to obtain a year recognition model;

[0016] The tangerine peel image to be identified is input into the year identification model to perform year prediction, and a year prediction result of the tangerine peel image to be identified is obtained.

[0017] In combination with the second aspect, in one embodiment of the present application, the tangerine peel year recognition module includes a first network module, which is used to extract multi-view features of the initial feature map; the first network module includes a deep convolution component, a layer normalization component, an activation function component, a point convolution component, a channel attention component and a spatial attention component.

[0018] In combination with the second aspect, in one embodiment of the present application, the tangerine peel year recognition module includes a second network module, which is used to identify the correlation features between different multi-view features; the second network module includes a pooling component, a fully connected component, an activation function component, a function component and a point convolution component.

[0019] On the other hand, an embodiment of the present application further provides an electronic device, including:

[0020] at least one processor;

[0021] at least one memory for storing at least one program;

[0022] When at least one of the programs is executed by at least one of the processors, the method for identifying the age of tangerine peel as described above is implemented.

[0023] On the other hand, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program executable by a processor, and the computer program executable by the processor is used to implement the above-mentioned method for identifying the age of tangerine peel when executed by the processor.

[0024] A method for identifying the year of dried tangerine peel provided in an embodiment of the present application first inputs a dried tangerine peel image marked with a year label into an initial year recognition model, and then extracts features from the input dried tangerine peel image to obtain an initial feature map. On this basis, the multi-view features in the initial feature map are further mined, and the deep associations between these features are accurately identified to fully capture the properties of dried tangerine peel under different viewing angles. Next, accurate year prediction is performed using the associated features as a basis, and the year prediction result is output. In order to continuously optimize the recognition performance, the prediction result is further compared with the known year label, and the parameters of the initial year recognition model are adjusted accordingly, and the optimized year recognition model is finally obtained. In practical applications, it is only necessary to input the dried tangerine peel image to be identified into the trained year recognition model to quickly obtain accurate year prediction results. The embodiment of the present application can more accurately capture the subtle differences of dried tangerine peel due to different years by automatically and more comprehensively capturing the multi-view features of dried tangerine peel and deeply analyzing the intrinsic connections between these features, thereby improving the accuracy and efficiency of dried tangerine peel year recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1It is a flow chart of the method for identifying the year of tangerine peel provided in an embodiment of the present application;

[0026] Figure 2 The embodiment of this application provides Figure 1 Schematic diagram of specific steps of step 130;

[0027] Figure 3 Another embodiment of the present application provides Figure 1 Schematic diagram of specific steps of step 130;

[0028] Figure 4 This is a diagram of a tangerine peel year identification system provided by an embodiment of the present application;

[0029] Figure 5 is a schematic diagram of the structure of the first network module provided in an embodiment of the present application;

[0030] Figure 6 is a schematic diagram of the structure of the second network module provided in an embodiment of the present application;

[0031] Figure 7 It is an architecture diagram of the year recognition model provided in the embodiment of the present application;

[0032] Figure 8 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0034] It should be noted that, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order from that in the flowchart. The terms "first", "second", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions that the present invention can be implemented, so they have no technical substantive significance. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" quoted in this specification are only for the convenience of narration, and are not used to limit the scope of the implementation of the present invention. The change or adjustment of the relative relationship should also be regarded as the scope of the implementation of the present invention without substantially changing the technical content.

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0036] In the traditional practice of tangerine peel year identification, people have long relied on manual experience and sensory evaluation to determine the year of tangerine peel. Although this method has a long history, it inevitably carries a strong subjective color and is easily interfered by human factors, which leads to inaccuracy and inconsistency in the results. Especially in the face of large-scale and batch detection needs, the efficiency bottleneck of manual identification is particularly prominent. It is not only time-consuming and laborious, but also requires comprehensive consideration of the front and back features of tangerine peel. The detection process is cumbersome and complicated, resulting in low detection efficiency. In addition, traditional methods are limited by high labor costs and specific equipment requirements, making it difficult for tangerine peel year identification to be flexible, efficient and accurate. It is worth noting that as an agricultural product, tangerine peel is deeply affected by natural environmental conditions such as light, temperature, and humidity. The subtle changes in these factors can leave traces on the characteristics of tangerine peel, further increasing the complexity and challenge of year identification. Therefore, directly applying a general network model to extract and classify the surface of tangerine peel often fails to accurately capture the subtle differences in tangerine peel due to different years, making the fault tolerance of the recognition result low and difficult to guarantee its accuracy and reliability.

[0037] In view of this, a method for identifying the year of dried tangerine peel, a system for identifying the year of dried tangerine peel, an electronic device and a computer-readable storage medium provided in an embodiment of the present application, the method first inputs a dried tangerine peel image marked with a year label into an initial year recognition model, and then extracts features from the input dried tangerine peel image to obtain an initial feature map. On this basis, the multi-view features in the initial feature map are further mined, and the deep associations between these features are accurately identified to fully capture the properties of dried tangerine peel at different viewing angles. Next, the associated features are used as a basis for accurate year prediction, and the year prediction results are output. In order to continuously optimize the recognition performance, the prediction results are further compared with the known year labels, and the parameters of the initial year recognition model are adjusted accordingly, and the optimized year recognition model is finally obtained. In practical applications, it is only necessary to input the dried tangerine peel image to be identified into the trained year recognition model to quickly obtain accurate year prediction results. The embodiment of the present application can more accurately capture the subtle differences of dried tangerine peel due to different years by automatically and more comprehensively capturing the features of dried tangerine peel at different viewing angles, and deeply analyzing the inherent connections between these features, thereby improving the accuracy and efficiency of dried tangerine peel year recognition.

[0038] The embodiments of the present application are further described below in conjunction with the accompanying drawings.

[0039] like Figure 1 As shown, Figure 1 1 is a flow chart of a method for identifying the age of dried tangerine peel provided in an embodiment of the present application. The method for identifying the age of dried tangerine peel includes but is not limited to steps 110 to 160.

[0040] Step 110: Obtain a tangerine peel image, input the tangerine peel image into an initial year recognition model, and set a year label for the tangerine peel image;

[0041] Step 120: extracting features from the tangerine peel image to obtain an initial feature map;

[0042] Step 130: extracting multi-view features of the tangerine peel image based on the initial feature map, and identifying correlation features between different multi-view features;

[0043] Step 140: performing year prediction according to the associated features to obtain a year prediction result;

[0044] Step 150: adjusting the parameters of the initial year recognition model according to the year prediction result and the year label to obtain the year recognition model;

[0045] Step 160: input the to-be-recognized tangerine peel image into a year recognition model to perform year prediction, and obtain a year prediction result of the to-be-recognized tangerine peel image.

[0046] Steps 110 to 160 are described in detail below.

[0047] In a feasible embodiment, the tangerine peel image refers to a clear image of the tangerine peel capsule and epidermis captured using a variety of image acquisition devices (including but not limited to smart phones, professional cameras, high-definition cameras, etc.), covering different years. These images are intended to show the key features of tangerine peel in detail, such as color, fineness of texture, shape contour, etc., to provide a solid foundation for subsequent in-depth analysis and accurate identification. In addition, the tangerine peel image will be set with a corresponding year label. Through the guidance of these labels, the model can more accurately capture the key features in the tangerine peel image, and in continuous iterative learning, gradually improve the accuracy and stability of year identification.

[0048] In a feasible embodiment, before the tangerine peel image is input into the initial year recognition model, the tangerine peel image can be preprocessed to improve the image quality and facilitate subsequent feature extraction and classification recognition. Specifically, the tangerine peel image is first classified according to the year to obtain multiple groups of tangerine peel image groups with different years, wherein each tangerine peel image group contains tangerine peel inner capsule image and tangerine peel epidermis image. Next, for each group of images, the color image is converted into a grayscale image to simplify the image information and focus on the brightness change; a filtering algorithm is used to remove the noise in the image, such as salt and pepper noise or Gaussian noise, to ensure the purity of the image; and the image is enhanced by histogram equalization, contrast enhancement and other methods to make the key features of the tangerine peel such as texture and wrinkles more distinct, which is convenient for model capture. After completing the above processing steps, the tangerine peel inner capsule image and tangerine peel epidermis image are annotated with the year according to the determined year classification standard. Further, the annotated tangerine peel images can be divided into a training set, a validation set and a test set according to a certain ratio. The training set is used to train the initial year recognition model; the validation set is used to adjust the model parameters; and the test set is used to evaluate the performance of the model. It is worth noting that when dividing the data set, it is necessary to ensure that the number of samples between categories is relatively balanced to avoid the problem of class imbalance.

[0049] In a feasible embodiment, the initial year recognition model can be a structure based on a deep learning model architecture (such as a convolutional neural network), which can at least include an input layer, a convolution layer, a global average pooling layer (GAP), a layer normalization layer, a fully connected layer, and an output layer, which can realize the year recognition of tangerine peel images. It should be noted that before the initial year recognition model is trained based on the tangerine peel image training set, the model can be pre-trained on a public data set (such as ImageNet). Specifically, ImageNet is a large visual database containing millions of pictures, covering thousands of categories. In the process of training the initial year recognition model using the ImageNet data set, the model can learn to extract useful features from the image, which can capture basic information such as color, texture, shape, and higher-level semantic information in the image. After the training is completed, the weight of the model is saved. These weights contain the knowledge and feature representation learned by the model on the ImageNet data set. Then, the pre-trained model weights can be used as a starting point to fine-tune the model to adapt to specific tasks, such as tangerine peel year recognition. During the fine-tuning process, the tangerine peel image dataset can be used directly to train the model. It is worth noting that at this time, training is no longer started from scratch, but fine-tuning is performed based on pre-trained weights, which helps speed up training and improve model performance.

[0050] In a feasible embodiment, before inputting the tangerine peel image training set into the initial year recognition model, in addition to the aforementioned preprocessing steps, the image size of the tangerine peel images can also be unified, for example, the tangerine peel images can be adjusted to a uniform 224x224 pixel size. This standardization operation helps the model maintain a consistent input dimension in subsequent processing.

[0051] In a feasible embodiment, in the process of extracting features from the tangerine peel image to obtain the initial feature map, the tangerine peel image can be first input into the backbone convolution layer (e.g., 4x4 backbone convolution layer) for convolution operation, which not only effectively captures the basic information in the image, but also significantly increases the number of channels of the feature map, providing rich data dimensions for subsequent deeper analysis. In order to ensure the stability and convergence of the model training and avoid the gradient explosion problem, a layer normalization layer is deployed after the backbone convolution layer. The layer normalization layer adjusts the feature map corresponding to each tangerine peel image to make the distribution of the feature map more stable, thereby effectively controlling the scale of the gradient and reducing the risk of gradient explosion, laying a solid foundation for the accuracy and robustness of the tangerine peel year recognition model.

[0052] In a feasible embodiment, multi-view features refer to a comprehensive feature set constructed by integrating image features from the same tangerine peel image but different views (such as surface view and internal capsule view, etc.), aiming to make full use of the complementarity of image information under different views to improve the accuracy and robustness of image recognition, classification or analysis. It is understandable that images under different views may contain different information, such as changes in lighting conditions may highlight different texture details, and changes in angles may reveal different sides of an object. Integrating this information helps to understand the image content more comprehensively. In addition, image features under a single view may be affected by factors such as noise, occlusion or lighting changes, while multi-view features can mitigate these effects by fusing information from multiple perspectives, thereby improving the robustness of the system. In addition, in practical applications, tangerine peel images may be affected by a variety of complex factors, such as background interference, shape variation, etc. Multi-view features can better adapt to these complex scenes and improve the performance of image processing. By integrating image features under different views, the accuracy of tangerine peel classification and recognition can be improved.

[0053] See also Figure 2 The specific process of extracting the multi-view features of the initial feature map in step 130 may include but is not limited to steps 210 to 240.

[0054] Step 210: extracting local features of the initial feature map, normalizing and feature mapping the local features, and obtaining a first feature map;

[0055] Step 220: performing channel attention calculation on the first feature map to obtain a second feature map;

[0056] Step 230: performing spatial attention calculation on the second feature map to obtain a third feature map;

[0057] Step 240: Aggregate the initial feature map and the third feature map to obtain a multi-view feature.

[0058] In a feasible embodiment, by extracting and normalizing local features from the initial feature map of the tangerine peel image, the comparability and stability between different features can be ensured. After all local features of the initial feature map are converted into the first feature map, a channel attention mechanism is introduced to dynamically adjust the first feature map, so that the model can focus on the channel information that is more critical to the recognition task.

[0059] After channel attention calculation, the second feature map is obtained, and its feature representation is optimized and enhanced in the channel dimension. Furthermore, spatial attention calculation is performed on the second feature map to capture the feature importance of different spatial positions in the image. The third feature map generated after spatial attention calculation realizes feature refinement and optimization in the spatial dimension. The initial feature map is aggregated with the third feature map enhanced by channel and spatial attention. This process not only retains the basic information in the original feature map, but also incorporates the feature representation optimized by the dual attention mechanism, which can effectively extract the low-level features (color, size, shape, etc.) and high-level features (texture) of tangerine peel itself, thereby realizing the comprehensive integration of multi-view features. The aggregated multi-view feature map contains rich local detail information and reflects the global semantic and structural features, providing strong support for subsequent tangerine peel image recognition, classification or analysis tasks.

[0060] In a feasible embodiment, the local feature X of the initial feature map i ∈R H×W×C After deep convolution operation, normalization and activation function processing, the corresponding first feature map X can be obtained. D The calculation process is shown in formula (1).

[0061]

[0062] Among them, δ is the ReLU activation function, μ is the layer normalization layer operation, and f 7×7 is a convolutional layer operation with a convolution kernel size of 7×7, R H×W×C Represents the feature map corresponding to the initial feature map, H represents height, W represents width, and C represents channel.

[0063] Furthermore, the first feature map X D Perform channel attention calculation to obtain the second feature map X C =F C (X D ), the calculation process is shown in formula (2).

[0064] F C (X D )=σ(W2δ(W1GAP(X D ))) (2),

[0065] Among them, F C ∈R 1×1×C , is the weight of the fully connected layer in the channel attention, δ is the ReLU activation function, and σ is the sigmoid activation function.

[0066] Furthermore, the second feature map X CPerform spatial attention calculation to obtain the third feature map X S =F S (X C ), the calculation process is shown in formula (3).

[0067]

[0068] Among them, F S ∈R H×W×1 , σ is the sigmoid function, * is the convolution operation, f 1×1 and f 3×3 They are convolutional layers with kernel sizes of 1×1 and 3×3 respectively.

[0069] Furthermore, the local feature X i With the third characteristic graph X S Aggregate and obtain the multi-view feature X through the ReLU activation function layer i+1 , the calculation process is shown in formula (4).

[0070] X i+1 =σ(X i +X S ) (4).

[0071] In a feasible embodiment, steps 210 to 240 may be repeated multiple times to obtain a multi-view feature with richer features. For example, steps 210 to 240 are repeated three times to obtain a primary multi-view feature, and then the primary multi-view feature is downsampled to prevent the loss of effective information. Then, the primary multi-view feature after the downsampling process is used as input to repeat steps 210 to 240 four times to obtain a secondary multi-view feature, and then the secondary multi-view feature is used as input to repeat steps 210 to 240 six times to obtain a tertiary multi-view feature, and then the tertiary multi-view feature is used as input to repeat steps 210 to 240 three times to obtain the final multi-view feature.

[0072] See also Figure 3 The specific process of identifying the associated features between different multi-view features in step 130 may include but is not limited to steps 310 to 330.

[0073] Step 310: compressing the multi-view features to obtain integrated features;

[0074] Step 320: performing average pooling, nonlinear combination and feature mapping processing on the integrated features to obtain a fourth feature map;

[0075] Step 330: perform feature integration on the initial feature map and the fourth feature map to obtain associated features.

[0076] In a feasible embodiment, since each group of tangerine peel images input includes a tangerine peel inner capsule image and a tangerine peel epidermis image, the multi-view features of the tangerine peel image extracted based on the initial feature map include tangerine peel epidermis image features and inner capsule image features. By compressing the epidermis image features and the inner capsule image features to obtain integrated features, the feature dimension can be reduced and redundant information can be removed.

[0077] It should be noted that average pooling refers to the average pooling operation of the integrated features to further reduce the spatial dimension of the features and enhance the robustness of the features. Average pooling helps to reduce the impact of noise and retain key information. Nonlinear combination refers to the nonlinear combination of the pooled features by introducing nonlinear activation functions (such as ReLU, Sigmoid, etc.) to enhance the expressiveness of the features. Nonlinear combination can capture the complex relationship between features and make the features more distinguishable. Feature mapping refers to the mapping of the features after nonlinear combination using a fully connected layer or a convolutional layer to obtain a new feature representation (i.e., the fourth feature map). It should be noted that the fourth feature map contains the features of two views of the tangerine peel image (i.e., tangerine peel epidermis image features and internal capsule image features). Step 320 can further refine and integrate feature information, emphasize the information of important views, and suppress irrelevant view information.

[0078] In a feasible embodiment, in step 330, the initial feature map and the fourth feature map are integrated (fused) to effectively integrate the information of different views to reveal the related features between them. The integration method can be simple concatenation, weighted summation, or a more complex attention mechanism.

[0079] In a feasible embodiment, the tangerine peel epidermis image feature f1 and the inner capsule image feature f2 are subjected to feature compression processing to obtain an integrated feature Y i ∈R C×H×W After average pooling, nonlinear combination (fully connected layer processing) and feature mapping processing, the corresponding fourth feature map X can be obtained. G The calculation process is shown in formula (5).

[0080] X G =σ(W2(δ(W1(GAP(X i )))) (5),

[0081] in, is the weight of the fully connected layer, and σ is the S igmo id function.

[0082] Furthermore, the integrated feature Y i With the fourth characteristic graph X G Multiply them together, and then perform feature integration by 1×1 convolution to obtain the associated feature Yi+1 The calculation process is shown in formula (6).

[0083]

[0084] Where Conv is a convolutional layer with a convolution kernel size of 1×1. is element-wise multiplication.

[0085] In a feasible embodiment, in the process of obtaining the year prediction result by performing year prediction according to the associated features, the associated features can be made decisions through the global average pooling layer, the fully connected layer and the Softmax function, and the year to which the tangerine peel image belongs is finally output. In this process, the associated features are first processed by the global average pooling layer to effectively reduce the parameter amount and enhance the spatial generalization ability of the model. Subsequently, these processed features are passed to the fully connected layer, which is responsible for further refining the feature information and learning the complex associations between the features, thereby providing a strong basis for year prediction. Finally, the output of the fully connected layer is normalized by the Softmax function, and this step ensures that the probability distribution sum of the output is 1, directly mapping the probability that the input tangerine peel image belongs to each year, thereby clearly outputting the year prediction result.

[0086] In a feasible embodiment, after completing the year prediction based on multi-view features and associated features, the parameters of the initial year recognition model can be adjusted by using optimization techniques such as back propagation algorithm based on the comparison between the prediction results and the actual year labels. This process aims to minimize the prediction error, improve the accuracy and generalization ability of the model, and thus obtain an optimized year recognition model. Once the year recognition model training is completed and its performance is verified to be up to standard, it can be applied to actual scenarios. The tangerine peel image to be identified is input into the model, and the model will automatically extract image features, analyze the association between features, and make year predictions based on the learned patterns. Finally, the model will output the year prediction results of the tangerine peel image to be identified, providing users with a fast and accurate basis for year judgment.

[0087] See also Figure 4 , Figure 4 It is a diagram of a tangerine peel year identification system provided in an embodiment of the present application, which can be applied to a tangerine peel year identification method as in any of the previous embodiments. The system includes: an image acquisition module 410 for acquiring a tangerine peel image, and a tangerine peel year identification module 420 for receiving a tangerine peel image and executing a tangerine peel year identification method as in any of the previous embodiments on the tangerine peel image.

[0088] In a feasible embodiment, the tangerine peel year recognition module 420 is equipped with a year recognition model, and can perform the tangerine peel year recognition method of any of the previous embodiments on the tangerine peel image. The year recognition model includes a first network module and a second network module. Figure 5 As shown, the first network module, as the core of feature extraction, integrates a deep convolution component (using a 7x7 depth-separable convolution layer, i.e., 7x7DWConv), a layer normalization component (layer normalization layer), two ReLU activation function components, a point convolution component (1x1 standard convolution layer, i.e., 1x1Conv), a channel attention component (cSE), and a spatial attention component (sSE). This module can accurately extract multi-view features from the initial feature map by performing a series of steps (steps 210 to 240), laying the foundation for subsequent associated feature recognition. Figure 6 As shown, the second network module focuses on identifying and integrating the deep associations between these multi-view features. The module consists of a pooling component (global average pooling layer GAP), a fully connected component (two fully connected layers), a ReLU activation function component, an additional point convolution component (for feature transformation), and a sigmoid function component. This architecture ensures that the second network module can efficiently execute steps 310 to 330, thereby accurately identifying and fusing the association information between multi-view features.

[0089] See also Figure 7 , Figure 7This is the architecture diagram of the year recognition model. The architecture integrates multiple key components to achieve efficient data processing and feature extraction. The core components of the architecture include an initial 4x4 standard convolution layer, which is responsible for initially capturing the basic features in the image. This is followed by a layer normalization layer, which is used to initially adjust the data distribution to ensure the stability of the training process. This is followed by a feature extraction layer, which is the hub of the entire architecture and is interwoven with multiple module groups and downsampling layers. Each downsampling layer consists of a layer normalization layer and a 2x2 convolution layer, which aims to extract higher-level features by reducing the resolution of the feature map while maintaining effective information transmission. It is worth noting that the output of each module group is directly connected to the downsampling layer. This design greatly promotes the flow of information and the multi-scale fusion of features, allowing the model to understand the image content more comprehensively. It should be noted that each module group is composed of multiple stacked first network modules. Specifically, the feature extraction layer includes four module groups and three downsampling layers, and the first module group includes three stacked first network modules, the second module group includes four stacked first network modules, the third module group includes six stacked first network modules, and the fourth module group includes three stacked first network modules, and each module group is connected to a downsampling layer. The close combination between each module group and the downsampling layer ensures the effective transmission of information between different levels and the multi-level extraction of features. After the feature extraction layer, the second network module continues to further refine and integrate the features. Subsequently, the pooling layer GAP effectively reduces the amount of calculation and extracts key information by reducing the dimension of the feature map. Next, the fully connected layer maps the feature vector output by the pooling layer to the target space. This process realizes the conversion from the feature space to the year prediction space. Finally, a sigmoid function component is used as the output layer to ensure the normalization of the prediction results, so that the output value can intuitively reflect the probability distribution of the tangerine peel year, thereby obtaining the year prediction result.

[0090] In a feasible embodiment, after the year recognition model is trained, the user only needs to input the tangerine peel image to be recognized into the year recognition model, and the model can automatically complete a series of complex operations such as feature extraction, feature association analysis, and year prediction. Finally, the model will output an accurate year prediction result, which provides great convenience for users.

[0091] In a feasible embodiment, the image acquisition module 410 and the tangerine peel year recognition module 420 can be integrated into various intelligent devices, especially the highly popular terminal device of smart phones, which greatly broadens its application scenarios and convenience. The image acquisition module 410 can utilize the built-in camera function and album access rights of the smart phone to realize the instant capture and import of tangerine peel images. Once the images are collected, they are seamlessly transmitted to the tangerine peel year recognition module 420, which then starts its efficient operation mechanism. Relying on the pre-trained year recognition model, the tangerine peel year recognition module 420 can quickly perform a deep analysis of the tangerine peel in the image. It uses advanced image recognition technology to accurately capture the subtle features of the texture, color, and morphology of the tangerine peel surface, and compares and analyzes with the massive data in the model, so as to quickly and accurately extract the year information of the tangerine peel. This integrated and intelligent design enables users to obtain accurate year information of tangerine peel anytime and anywhere.

[0092] Reference Figure 8 The embodiment of the present application further discloses an electronic device, the electronic device 800 comprising:

[0093] at least one processor 801;

[0094] At least one memory 802, used to store at least one program;

[0095] When at least one program is executed by at least one processor 801, the aforementioned tangerine peel year identification method is implemented.

[0096] The embodiment of the present application also discloses a computer-readable storage medium, which stores a computer program executable by a processor. When the computer program executable by the processor is executed by the processor, it is used to implement the tangerine peel year identification method as described above.

[0097] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying the year of tangerine peel, characterized in that: include: Acquire a tangerine peel image, input the tangerine peel image into an initial year recognition model, and set a year label for the tangerine peel image; Extracting features from the tangerine peel image to obtain an initial feature map; Extracting multi-view features of the initial feature map, and identifying correlation features between different multi-view features; Perform year prediction according to the correlation characteristics to obtain year prediction results; Adjusting the parameters of the initial year recognition model according to the year prediction result and the year label to obtain a year recognition model; The tangerine peel image to be identified is input into the year identification model to perform year prediction, and a year prediction result of the tangerine peel image to be identified is obtained.

2. The method for identifying the age of dried tangerine peel according to claim 1, characterized in that: The extracting multi-view features of the initial feature map includes: extracting local features of the initial feature map, normalizing and feature mapping the local features to obtain a first feature map; performing channel attention calculation on the first feature map to obtain a second feature map; performing spatial attention calculation on the second feature map to obtain a third feature map; and aggregating the initial feature map and the third feature map to obtain multi-view features.

3. The method for identifying the age of dried tangerine peel according to claim 1, characterized in that: The identifying of the associated features between the different multi-view features includes: performing feature compression processing on the multi-view features to obtain integrated features; performing average pooling, nonlinear combination and feature mapping processing on the integrated features to obtain a fourth feature map; and performing feature integration on the initial feature map and the fourth feature map to obtain associated features.

4. The method for identifying the age of dried tangerine peel according to claim 3, characterized in that: The multi-view features include a front view feature and a back view feature of the tangerine peel image; The step of performing feature compression processing on the multi-view features to obtain an integrated feature includes: compressing the front view feature and the back view feature of the tangerine peel image into a single feature to obtain an integrated feature.

5. The method for identifying the age of dried tangerine peel according to claim 1, characterized in that: Before the tangerine peel image is input into the initial year recognition model, the method includes: denoising the tangerine peel image; and unifying the image size of the tangerine peel image.

6. A system for identifying the age of tangerine peel, characterized in that: include: An image acquisition module, used for acquiring tangerine peel images; The tangerine peel year recognition module is used to receive the tangerine peel image and perform the following tangerine peel year recognition method on the tangerine peel image: Inputting the tangerine peel image into an initial year recognition model, wherein the tangerine peel image is provided with a year label; Extracting features from the tangerine peel image to obtain an initial feature map; Extracting multi-view features of the initial feature map, identifying associated features between different multi-view features, the multi-view features including front view features and back view features of the tangerine peel image; Perform year prediction according to the correlation characteristics to obtain year prediction results; Adjusting the parameters of the initial year recognition model according to the year prediction result and the year label to obtain a year recognition model; The tangerine peel image to be identified is input into the year identification model to perform year prediction, and a year prediction result of the tangerine peel image to be identified is obtained.

7. The tangerine peel year identification system according to claim 6, characterized in that: The tangerine peel year recognition module includes a first network module, which is used to extract multi-view features of the initial feature map; the first network module includes a deep convolution component, a layer normalization component, an activation function component, a point convolution component, a channel attention component and a spatial attention component.

8. The tangerine peel year identification system according to claim 6, characterized in that: The tangerine peel year recognition module includes a second network module, which is used to identify the associated features between different multi-view features; the second network module includes a pooling component, a fully connected component, an activation function component, a function component and a point convolution component.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, the method for identifying the age of tangerine peel as described in any one of claims 1 to 5 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program executable by a processor is stored therein, and when the computer program executable by the processor is executed by the processor, it is used to implement the method for identifying the year of tangerine peel as described in any one of claims 1 to 5.

Images