Automatic sky bead image recognition method and system based on improved convolutional neural network

Abstract

The invention provides an improved convolutional neural network-based sky-bead image automatic identification method and system, and the method comprises the steps: collecting and preprocessing sky-bead images with different patterns and shapes, carrying out the data augmentation operation of the sky-bead images, and building a training set, a verification set and a test set which are needed by a sky-bead image identification model; constructing a convolutional neural network model based on improved OfficientNetV2, introducing a new activation function and an attention mechanism, loading training and verification sets in batches to carry out iterative training on a verification model, and continuously adjusting network parameters to obtain an optimal sky bead image recognition model; inputting a test sky bead image to the optimal model, and returning a sky bead image automatic identification result; according to the method provided by the invention, the accuracy and reasoning speed of sky bead image recognition can be improved on the premise of reducing calculation parameters, a lightweight model can be built for application, the problems of low accuracy, time and labor consumption and the like of traditional manual recognition are solved, and rapid and accurate recognition of the sky bead image can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and pattern recognition, and in particular to a method and system for automatically recognizing dzi bead images based on an improved convolutional neural network. Background Art

[0002] Dzi beads are a special type of jewelry, mainly made of corroded agate. They come in long, short, round and flat shapes with complex and diverse patterns. These patterns not only reflect different meanings and values, but also attract the attention of many Dzi enthusiasts and experts in the field. The pattern characteristics of Dzi beads are the main basis for their image recognition. However, traditional Dzi bead recognition methods mainly rely on manual observation and identification, which is not only time-consuming and labor-intensive, but also inefficient, and the recognition accuracy is often affected by the subjective experience of the identifier. In addition, the differences in the age, material and color of Dzi beads further increase the complexity and difficulty of recognition.

[0003] With the rapid development of deep learning technology, the performance of image classification tasks based on convolutional neural networks has been significantly improved, and has been successfully applied to image recognition in multiple fields. The effect of deep learning methods in image classification and recognition is far superior to traditional manual identification technology. Its advantage is that it does not rely on the subjective experience of experts and manually designed features. By inputting domain images into the deep learning model, the model can automatically perform preprocessing and feature extraction, and return the label of the category to which the image belongs.

[0004] The general steps of image classification and recognition based on deep learning include: first, construct an image classification dataset and divide it into training set, validation set and test set in a certain proportion; second, input the training validation image dataset into the model for iterative training, continuously optimize the model to extract the significant features of images of different categories, and adjust the hyperparameters of the model to obtain the optimal image classification model and save it; finally, input the test set into the optimal model and return the predicted category result to evaluate the performance of the image classification model.

[0005] In recent years, image classification methods based on deep learning have made breakthrough developments, especially in the field of convolutional neural networks. Since the convolutional neural network AlexNet performed well in the ImageNet image recognition challenge in 2012, models based on deep networks such as GoogleNet, ResNet and DenseNet have been proposed one after another. These models have achieved better fitting ability and classification effect by deepening the number of network layers. However, as the number of network layers increases, the amount of parameter calculation of the model also increases, and the inference speed slows down, which affects the efficiency of the model in practical applications.

[0006] To address this problem, researchers have drawn on their respective strengths and proposed lightweight deep neural networks, such as MobileNet, ShuffleNet, EfficientNet, and Vision Transformer. These models can improve image classification performance while reducing the amount of parameter calculation. Among them, EfficientNetV2 incorporates an adaptive regularization strategy into its design. Its verification results on large-scale image datasets such as ImageNet and CIFAR show that compared with other models, the accuracy rate is improved by 2%, the training speed is accelerated by 5 to 11 times, and the number of calculated parameters is reduced by 6.8 times, providing a good foundation for the lightweight deployment of image classification models. To this end, the present invention further improves and optimizes the EfficientNetV2 basic model to meet the application requirements of a fast and accurate automatic recognition model for dzi beads images, designs a more lightweight dzi bead recognition model, and introduces a new Hardswish activation function and CAA attention mechanism through model improvement to improve the accuracy and reasoning speed of the recognition model while reducing the amount of calculation, effectively solving the problems of low work efficiency, high cost and accuracy dependence on the subjective experience of identification experts in the current traditional manual identification of dzi beads, and can realize the lightweight deployment, application and promotion of the model, providing efficient and intelligent auxiliary role in the identification of traditional dzi beads. Summary of the invention

[0007] In view of the defects of the prior art, the present invention provides a method and system for automatic recognition of Dzi bead images based on an improved convolutional neural network. The Dzi bead image data obtained by different acquisition methods are input into the improved convolutional neural network model for automatic preprocessing, extraction of significant features and classification recognition, which reduces the calculation of model parameters while accelerating model fitting, improving the training and reasoning speed of the model and the recognition accuracy, realizing automatic recognition of Dzi bead images, and laying a foundation for lightweight application deployment and auxiliary identification.

[0008] In order to achieve the above invention object, the technical solution adopted by the present invention is as follows:

[0009] A method for automatically identifying dzi beads images based on an improved convolutional neural network, the method comprising the following steps:

[0010] Collecting Dzi bead images: Using video equipment to take photos of Dzi bead markets, collecting images of different categories of Dzi beads, and combining expert field knowledge and reference materials, using web crawling to collect Dzi bead images with different lighting angles, color backgrounds, and shapes, to form 23 categories of Dzi bead patterns;

[0011] Establish a Dzi bead image dataset: pre-process the collected Dzi bead images, and expand the Dzi bead image dataset by combining data enhancement operations, and annotate the Dzi bead image data with different pattern features using the lowercase combination of the pinyin first letters of the Dzi bead names;

[0012] Construct an improved convolutional neural network model: The EfficientNetV2 model based on a deep convolutional neural network is used for improvement. In the feature extraction stages 1 to 6, the CAA-Fused-MBConv module is used for shallow feature extraction in stages 1 to 3, and the CAA-MBConv module is used for high-level semantic feature extraction in stages 4 to 6. The Hardswish activation function is used instead of the Swish activation function, and the SE module is changed to the CAA attention mechanism to reduce the model calculation complexity, speed up the model training and reasoning speed, and improve the model fitting ability and recognition accuracy.

[0013] Automatic recognition of Dzi bead images: Input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction result of the Dzi bead image, that is, the accuracy of the Dzi bead image and its pattern category.

[0014] Furthermore, the network crawling method uses Python's BeautifulSoup crawler library to crawl Baidu Gallery, Google Gallery, Bing Gallery and Dzi Bead website.

[0015] Furthermore, the categories of dzi bead patterns include: vajra dzi beads, water pattern dzi beads, swastika dzi beads, eight trigrams dzi beads, gold wire hook dzi beads, peace cross dzi beads, god of wealth dzi beads, Guanyin dzi beads, vase dzi beads, noble people dzi beads, tortoise shell dzi beads, tiger tooth dzi beads, wishful thinking dzi beads, dragon eye dzi beads, bodhi dzi beads, happy Buddha dzi beads, lotus magic instrument dzi beads, heaven and earth dzi beads, phoenix eye dzi beads, lightning five-eyed dzi beads, medicine master dzi beads, sun, moon and star dzi beads and lotus dzi beads.

[0016] Furthermore, the preprocessing steps for establishing the Dzi bead image dataset include eliminating repeated, irrelevant and unclear Dzi bead images; the data enhancement operations include performing Gaussian noise, cropping, rotation and color dithering on the collected Dzi bead images; the Dzi bead image data labeled with different pattern features is to divide the labeled dataset into training set, verification set and test set in a ratio of 8:2:1.

[0017] Furthermore, the step of constructing an improved convolutional neural network model includes: inputting the set batch size into the training verification data into the improved convolutional neural network model, efficiently extracting the significant features of the Dzi bead image by improving the activation function and attention mechanism, and realizing the automatic recognition of the Dzi bead image; inputting the Dzi bead image test data into the trained and saved model, testing the model fitting generalization performance, and continuously adjusting the parameters to obtain the final optimized Dzi bead image recognition model.

[0018] Furthermore, the notable features of the Dzi bead image include: Dzi bead patterns of different colors, shapes, sizes and backgrounds.

[0019] The present invention also discloses a Dzi bead image automatic recognition system based on an improved convolutional neural network. The system can be used to implement the above-mentioned Dzi bead image automatic recognition method based on an improved convolutional neural network, which specifically includes:

[0020] Image acquisition module: Use video equipment to take on-site photos of the Dzi bead market, collect images of different categories of Dzi beads, and combine expert field knowledge and reference materials to use network crawling to collect Dzi bead images with different lighting angles, color backgrounds and shapes to form 23 categories of Dzi bead patterns.

[0021] Data preprocessing module: preprocess the collected Dzi bead images, including removing duplicate and irrelevant images and performing blur processing.

[0022] Data enhancement module: Combine data enhancement operations to expand the Dzi bead image dataset, and improve the generalization ability of the model through Gaussian noise, cropping, rotation, and color jitter.

[0023] Label annotation module: annotate Dzi bead images and use the lowercase initials of the Dzi bead names to identify Dzi bead image data with different pattern features.

[0024] Model building module: The EfficientNetV2 model based on the deep convolutional neural network is improved, the CAA-Fused-MBConv and CAA-MBConv modules are used for feature extraction, and the Hardswish activation function and CAA attention mechanism are used to optimize the model structure to form an improved convolutional neural network model.

[0025] Model training module: Train and verify the improved convolutional neural network model to optimize fitting capabilities and improve recognition accuracy.

[0026] Image recognition module: Input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction results of the image and its pattern category.

[0027] Result output module: displays and saves the prediction results of Dzi bead images, including accuracy and category information.

[0028] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned automatic recognition method of Dzi beads images based on the improved convolutional neural network is implemented.

[0029] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned automatic recognition method of Dzi beads images based on the improved convolutional neural network is implemented.

[0030] Compared with the prior art, the advantages of the present invention are:

[0031] 1. Improve recognition accuracy: By improving the EfficientNetV2 model and introducing the Hardswish activation function and CAA attention mechanism, the accuracy of image recognition is significantly improved, and the number of misidentifications and missed identifications is reduced.

[0032] 2. Reduce computational complexity: The optimized network structure effectively reduces the computational complexity of the model, improves the reasoning speed of the model, and makes the recognition process more efficient and suitable for real-time applications.

[0033] 3. High degree of automation: This method no longer relies on manual feature design. The system can automatically preprocess and extract image features, reducing the need for manual intervention and improving work efficiency.

[0034] 4. Strong adaptability: Through data enhancement technology, the model has stronger generalization ability and can adapt to Dzi bead images with different lighting, backgrounds and shapes, enhancing the reliability of recognition.

[0035] 5. Lightweight deployment: The improved model is suitable for lightweight deployment, which facilitates efficient image recognition on mobile devices or edge computing devices and expands the application scenarios.

[0036] 6. Intelligent assistance function: It provides intelligent assistance for the identification of traditional Dzi beads, reduces the workload of identification experts, improves recognition efficiency and accuracy, and meets the market demand for efficient identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of the automatic recognition method of Dzi beads images according to an embodiment of the present invention;

[0038] Figure 2 exemplary images of Dzi beads of different pattern categories according to the embodiment of the present invention;

[0039] Figure 3 is a structural diagram of an improved convolutional neural network model constructed in an embodiment of the present invention;

[0040] Figure 4 This is a CAA attention structure diagram of an embodiment of the present invention;

[0041] Figure 5 It is a curve diagram of the verification accuracy of the automatic recognition model of Dzi beads images according to the embodiment of the present invention;

[0042] Figure 6 It is a verification loss curve diagram of the automatic recognition model of Dzi beads images according to an embodiment of the present invention;

[0043] Figure 72 are some example diagrams of the results of Dzi bead image recognition in the embodiments of the present invention; wherein (a) is the probability that the predicted image is an image of Bagua Dzi bead, which is 0.991; (b) is the probability that the predicted image is an image of God of Wealth Dzi bead, which is 0.986; (c) is the probability that the predicted image is an image of Lotus Dharma Instrument Dzi bead, which is 0.967; (d) is the probability that the predicted image is an image of Gold Wire Hook Dzi bead, which is 0.983; (e) is the probability that the predicted image is an image of Wan Zi Dzi bead, which is 0.976; and (f) is the probability that the predicted image is an image of Guanyin Dzi bead, which is 0.981. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

[0045] A method for automatic recognition of Dzi beads images based on an improved convolutional neural network. The specific process is as follows: Figure 1 As shown, the following steps are included:

[0046] Constructing Dzi bead image dataset: The Dzi bead image dataset used is firstly obtained by taking Dzi bead images of different patterns in different viewing angles, light intensities and complex backgrounds in the Dzi bead market with the help of different models of digital cameras and smartphone cameras to ensure the quality and diversity of Dzi bead images. Secondly, the Python crawling framework is used to obtain Dzi bead image data with different color backgrounds and pattern shape features from Baidu Gallery, Google Gallery, Bing Gallery and Dzi bead websites. Finally, based on the knowledge of experts in the field and Dzi bead works, the dataset is finally formed. Figure 2 The 23 different types of Dzi beads images shown in the figure include Vajra Dzi, Water Ripple Dzi, Wan Zi Dzi, Bagua Dzi, Gold Hook Dzi, Peace Cross Dzi, God of Wealth Dzi, Guanyin Dzi, Aquarius Dzi, Nobleman Dzi, Tortoise Shell Dzi, Tiger Tooth Dzi, Ruyi Dzi, Dragon Eye Dzi, Bodhi Dzi, Happy Buddha Dzi, Lotus Instrument Dzi, Heaven and Earth Dzi, Phoenix Eye Dzi, Lightning Five Eyed Dzi, Medicine Master Dzi, Sun, Moon and Star Dzi and Lotus Dzi. In addition, the collected image data are preprocessed by deleting irrelevant, duplicated and blurred images to reduce noise data, and the Dzi image dataset is effectively expanded using data enhancement strategies such as random Gaussian noise, cropping, rotation and color jittering. Finally, a dataset of 24610 Dzi images of different categories, Dzipattern, is formed to effectively avoid overfitting of the model and accelerate the rapid convergence of the model. The dataset is randomly divided into training, validation and test sets in a ratio of 8:2:1 to further train, validate and evaluate the performance of the model.

[0047] Design and improve the convolutional neural network model: Use the EfficientNetV2 image recognition model based on a deep convolutional neural network as the basic model to design an improved convolutional neural network model such as Figure 3 As shown in FIG, the model structure includes seven stages, of which the feature extraction stages 1 to 6 are CAA-Fused-MBConv modules with expansion factors not equal to 1 for shallow feature extraction, and stages 4 to 6 are CAA-MBConv modules for high-level semantic feature extraction. Among them, the CAA-Fused-MBConv module is composed of Conv3x3 convolution, introduced CAA attention and Conv1x1 convolution, and the CAA-MBConv module includes Conv1x1 convolution, Conv3x3 depth-separable convolution, introduced CAA attention and Conv1x1 convolution. The module uses Hardswish to replace the original Swish activation function, and uses piecewise linear calculation to simplify the computational complexity of the Swish function, so as to achieve better performance while reducing the amount of parameter calculation. In addition, the CAA (Context Anchor Attention) context anchor attention in the above module is improved and introduced on the basis of eliminating SE with excessively large global calculation parameters. The CAA structure is as follows Figure 4 As shown in the figure, using global average pooling and one-dimensional strip convolution to enhance the target area features can effectively capture the contextual information around the target object, which helps to deeply and accurately understand the target object in the image, thereby improving the model fitting ability and recognition accuracy.

[0048] Automatic recognition of Dzi bead images: By inputting the training set and the validation set into the constructed improved convolutional neural network model, the Dzi bead image recognition model is iteratively trained and verified. 150 rounds of iterative training are set to verify the model. The performance of the model is verified after every few rounds of iterative training. The iterative weight parameters are continuously updated using the back propagation and gradient descent methods until the model converges and the optimal model is saved. Among them, the accuracy fitting curve output by the validation set during the iterative training and verification model is as follows: Figure 5 As shown in FIG. 1 , the horizontal axis is the iteration round, the vertical axis is the verification accuracy, the dark curve is the verification accuracy curve obtained by the method proposed in the present invention, the light curve is the verification accuracy curve obtained by the EfficientNetV2 baseline model, and the difference loss fitting curve between the value calculated and output by the iterative verification model and the true value is shown in FIG. Figure 6 As shown in the figure, the horizontal axis is the iteration round, the vertical axis is the loss value, the dark loss curve is the loss fitting curve obtained by the method proposed in the present invention, and the light loss curve is the loss fitting curve obtained by the baseline model; in addition, the prediction results returned by the input test Dzi bead image using the optimal Dzi bead image recognition model saved after iterative training verification are as follows Figure 7As shown, among them, Figure (a) is the probability of predicting that the image is an image of Bagua Dzi Beads, which is 0.991, Figure (b) is the probability of predicting that the image is an image of God of Wealth Dzi Beads, which is 0.986, Figure (c) is the probability of predicting that the image is an image of Lotus Dharma Instrument Dzi Beads, which is 0.967, Figure (d) is the probability of predicting that the image is an image of Gold Wire Hook Dzi Beads, which is 0.983, Figure (e) is the probability of predicting that the image is an image of Wan Zi Dzi Beads, which is 0.976, and Figure (f) is the probability of predicting that the image is an image of Guanyin Dzi Beads, which is 0.981. The average verification accuracy of the method proposed in the present invention in each pattern category is 96.38%, and the inference speed is 0.039s, which effectively realizes the lightweight deployment application of the model and the rapid and accurate recognition of Dzi Bead images, solving the problems of low efficiency, low recognition accuracy and high cost faced by the prior art.

[0049] In another embodiment of the present invention, a system for automatically identifying Dzi beads images based on an improved convolutional neural network is provided. The system can be used to implement the above-mentioned method for automatically identifying Dzi beads images based on an improved convolutional neural network, specifically comprising:

[0050] Image acquisition module: Use video equipment to take on-site photos of the Dzi bead market, collect images of different categories of Dzi beads, and combine expert field knowledge and reference materials to use network crawling to collect Dzi bead images with different lighting angles, color backgrounds and shapes to form 23 categories of Dzi bead patterns.

[0051] Data preprocessing module: preprocess the collected Dzi bead images, including removing duplicate and irrelevant images and performing blur processing.

[0052] Data enhancement module: Combine data enhancement operations to expand the Dzi bead image dataset, and improve the generalization ability of the model through Gaussian noise, cropping, rotation, and color jitter.

[0053] Label annotation module: annotate Dzi bead images and use the lowercase initials of the Dzi bead names to identify Dzi bead image data with different pattern features.

[0054] Model building module: The EfficientNetV2 model based on the deep convolutional neural network is improved, the CAA-Fused-MBConv and CAA-MBConv modules are used for feature extraction, and the Hardswish activation function and CAA attention mechanism are used to optimize the model structure.

[0055] Model training module: Train and verify the improved convolutional neural network model to optimize fitting capabilities and improve recognition accuracy.

[0056] Image recognition module: Input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction results of the image and its pattern category.

[0057] Result output module: displays and saves the prediction results of Dzi bead images, including accuracy and category information.

[0058] In another embodiment of the present invention, a terminal device is provided, the terminal device includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the automatic recognition method of Dzi beads image, including the following steps:

[0059] Collecting Dzi bead images: Using video equipment to take photos of Dzi bead markets, collecting images of different categories of Dzi beads, and combining expert field knowledge and reference materials, using web crawling to collect Dzi bead images with different lighting angles, color backgrounds, and shapes, to form 23 categories of Dzi bead patterns;

[0060] Establish a Dzi bead image dataset: pre-process the collected Dzi bead images, and expand the Dzi bead image dataset by combining data enhancement operations, and annotate the Dzi bead image data with different pattern features using the lowercase combination of the pinyin first letters of the Dzi bead names;

[0061] Construct an improved convolutional neural network model: The EfficientNetV2 model based on a deep convolutional neural network is used for improvement. In the feature extraction stages 1 to 6, the CAA-Fused-MBConv module is used for shallow feature extraction in stages 1 to 3, and the CAA-MBConv module is used for high-level semantic feature extraction in stages 4 to 6. The Hardswish activation function is used instead of the Swish activation function, and the SE module is changed to the CAA attention mechanism.

[0062] Automatic recognition of Dzi bead images: Input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction result of the Dzi bead image, that is, the accuracy of the Dzi bead image and its pattern category.

[0063] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage.

[0064] The processor may load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of automatic recognition of Dzi bead images in the above-mentioned embodiment; the processor may load and execute the following steps:

[0065] Collecting Dzi bead images: Using video equipment to take photos of Dzi bead markets, collecting images of different categories of Dzi beads, and combining expert field knowledge and reference materials, using web crawling to collect Dzi bead images with different lighting angles, color backgrounds, and shapes, to form 23 categories of Dzi bead patterns;

[0066] Establish a Dzi bead image dataset: pre-process the collected Dzi bead images, and expand the Dzi bead image dataset by combining data enhancement operations, and annotate the Dzi bead image data with different pattern features using the lowercase combination of the pinyin first letters of the Dzi bead names;

[0067] Construct an improved convolutional neural network model: The EfficientNetV2 model based on a deep convolutional neural network is used for improvement. In the feature extraction stages 1 to 6, the CAA-Fused-MBConv module is used for shallow feature extraction in stages 1 to 3, and the CAA-MBConv module is used for high-level semantic feature extraction in stages 4 to 6. The Hardswish activation function is used instead of the Swish activation function, and the SE module is changed to the CAA attention mechanism.

[0068] Automatic recognition of Dzi bead images: Input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction result of the Dzi bead image, that is, the accuracy of the Dzi bead image and its pattern category.

[0069] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0070] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0071] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0072] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0073] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and should be understood that the protection scope of the present invention is not limited to such special statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for automatic recognition of Dzi beads images based on an improved convolutional neural network, characterized in that: The following steps are involved: Collecting Dzi bead images: Using video equipment to take photos of Dzi bead markets, collect images of different categories of Dzi beads, and use web crawling to collect images of Dzi beads with different lighting angles, color backgrounds, and shapes, in combination with expert domain knowledge and reference materials, to form Dzi bead pattern categories; Establish a Dzi bead image dataset: pre-process the collected Dzi bead images, and expand the Dzi bead image dataset by combining data enhancement operations, and annotate the Dzi bead image data with different pattern features using the lowercase combination of the pinyin first letters of the Dzi bead names; Construct an improved convolutional neural network model: The EfficientNetV2 model based on a deep convolutional neural network is used for improvement. In the feature extraction stages 1 to 6, the CAA-Fused-MBConv module is used for shallow feature extraction in stages 1 to 3, and the CAA-MBConv module is used for high-level semantic feature extraction in stages 4 to 6. The Hardswish activation function is used instead of the Swish activation function, and the SE module is changed to the CAA attention mechanism. Automatic recognition of Dzi bead images: Input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction result of the Dzi bead image, that is, the accuracy of the Dzi bead image and its pattern category.

2. The automatic identification method of Dzi beads according to claim 1, characterized in that: The network crawling method uses Python's BeautifulSoup crawler library to crawl Baidu Gallery, Google Gallery, Bing Gallery and Dzi Bead website.

3. The automatic identification method of Dzi beads according to claim 1, characterized in that: The categories of dzi patterns include: Vajra dzi beads, water pattern dzi beads, swastika dzi beads, eight trigrams dzi beads, gold wire hook dzi beads, peace cross dzi beads, god of wealth dzi beads, Guanyin dzi beads, vase dzi beads, noble people dzi beads, tortoise shell dzi beads, tiger tooth dzi beads, wishful thinking dzi beads, dragon eye dzi beads, bodhi dzi beads, happy Buddha dzi beads, lotus magic instrument dzi beads, heaven and earth dzi beads, phoenix eye dzi beads, lightning five-eyed dzi beads, medicine master dzi beads, sun, moon and star dzi beads and lotus dzi beads.

4. The automatic identification method of Dzi beads according to claim 1, characterized in that: The preprocessing steps for establishing the Dzi bead image dataset include eliminating repeated, irrelevant and unclear Dzi bead images; the data enhancement operation includes Gaussian noise, cropping, rotation and color dithering of the collected Dzi bead images; the Dzi bead image data labeled with different pattern features is to divide the labeled dataset into training set, verification set and test set in a ratio of 8:2:

1.

5. The automatic identification method of Dzi beads according to claim 1, characterized in that: The steps of constructing an improved convolutional neural network model include: inputting the set batch size into the training verification data into the improved convolutional neural network model, efficiently extracting the significant features of the Dzi bead image by improving the activation function and attention mechanism, and realizing the automatic recognition of the Dzi bead image; inputting the Dzi bead image test data into the trained and saved model, testing the model fitting generalization performance, and continuously adjusting the parameters to obtain the final optimized Dzi bead image recognition model.

6. The automatic identification method of Dzi beads according to claim 1, characterized in that: The notable features of the Dzi bead images include: Dzi bead patterns of different colors, shapes, sizes and backgrounds.

7. An automatic Dzi bead image recognition system based on an improved convolutional neural network, characterized in that: The system can be used to implement the automatic recognition method of Dzi beads images based on the improved convolutional neural network as described in any one of claims 1 to 6, specifically comprising: Image acquisition module: Use camera equipment to take photos of Dzi bead markets, collect images of different categories of Dzi beads, and use web crawling to collect images of Dzi beads with different lighting angles, color backgrounds, and shapes in combination with expert domain knowledge and reference materials to form Dzi bead pattern categories; Data preprocessing module: preprocess the collected Dzi bead images, including removing duplicate and irrelevant images and performing blur processing; Data enhancement module: Combine data enhancement operations to expand the Dzi bead image dataset, and improve the generalization ability of the model through Gaussian noise, cropping, rotation, and color jittering; Labeling module: labeling Dzi bead images, using the lowercase combination of the pinyin initials of the Dzi bead names to identify Dzi bead image data with different pattern features; Model building module: Improve the EfficientNetV2 model based on the deep convolutional neural network, use CAA-Fused-MBConv and CAA-MBConv modules for feature extraction, use Hardswish activation function and CAA attention mechanism to optimize the model structure, and form an improved convolutional neural network model; Model training module: train and verify the improved convolutional neural network model to optimize the fitting ability and improve the recognition accuracy; Image recognition module: input the Dzi bead image to be tested into the improved convolutional neural network model and return the prediction results of the image and its pattern category; Result output module: displays and saves the prediction results of Dzi bead images, including accuracy and category information.

8. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for automatically identifying images of Dzi beads as claimed in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the method for automatically identifying Dzi beads images as described in one of claims 1 to 6 is implemented.

Images