Image recognition system based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an image recognition system based on artificial intelligence. According to the system, in the data preprocessing stage, the advanced denoising technology is adopted, the image quality is remarkably improved, the negative influence of data noise on the classification and recognition accuracy is reduced, the consistency of the image in the feature extraction process is ensured by calculating the noise level of the denoised image and generating a normalized color histogram, and the classification and recognition accuracy is improved. The feature extraction unit generates feature vectors by using a convolutional neural network and performs systematic evaluation and normalization on the feature vectors, so that features of different images can be effectively compared, and the efficiency and accuracy of feature extraction are improved, the classification and recognition unit analyzes the feature vectors and calculates a confusion matrix through a decision tree algorithm, and the classification and recognition efficiency is improved. The reliability of the classification result is further ensured, and the series of improvements not only improve the overall performance and reliability of the image recognition system, but also provide more accurate image information output in practical application.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly to an image recognition system based on artificial intelligence. Background Art

[0002] With the rapid development of artificial intelligence technology, image recognition systems have been widely applied in various fields, such as medical image analysis, intelligent monitoring, autonomous driving, and social media content management. These systems can greatly improve work efficiency, reduce labor costs, and enhance the accuracy of data processing through automatic analysis of image data. In addition, artificial intelligence image recognition systems can identify and classify image content in real time, providing users with more intuitive and valuable information to help optimize decision-making and actions.

[0003] However, existing image recognition technologies often face problems such as data noise, inconsistent image quality, and low feature extraction efficiency when dealing with a large amount of image data. Specifically, many image recognition systems fail to effectively remove noise in images during the data preprocessing and feature extraction stages, resulting in a decrease in the accuracy of classification and recognition. In addition, existing technologies often lack a mechanism for systematically evaluating and normalizing feature vectors, making it difficult to effectively compare the results of feature extraction between different images. These technical problems limit the performance and reliability of image recognition systems in practical applications. Summary of the Invention

[0004] (I) Technical Problems to be Solved In view of the deficiencies of the prior art, the present invention provides an image recognition system based on artificial intelligence. In the data preprocessing stage, advanced denoising technology is adopted to significantly improve the image quality, thereby reducing the negative impact of data noise on the accuracy of classification and recognition. By calculating the noise level of the denoised image and generating a normalized color histogram, the consistency of the image during the feature extraction process is ensured. The feature extraction unit uses a convolutional neural network to generate feature vectors and systematically evaluates and normalizes them, enabling effective comparison of features between different images, thereby improving the efficiency and accuracy of feature extraction. The classification and recognition unit analyzes the feature vectors through a decision tree algorithm and calculates a confusion matrix to further ensure the reliability of the classification results. This series of improvements not only enhances the overall performance and reliability of the image recognition system but also provides more accurate image information output in practical applications, solving the above problems.

[0005] (II) Technical Solutions To achieve the above object, the present invention provides the following technical solution: An image recognition system based on artificial intelligence, comprising a data acquisition unit, a data preprocessing unit, a feature extraction unit, a classification and recognition unit, and a visualization output unit; The data acquisition unit batch collects images from the image library and records the metadata of each image, including the shooting time, location, and shooting device information; The data preprocessing unit extracts all the images from the data acquisition unit, adjusts the size of all the images, removes the noise in all the images using denoising techniques, calculates the noise level of the denoised images, calculates the color histogram of the images, generates the normalized color histogram data, and outputs all the enhanced images to the feature extraction unit; The feature extraction unit receives the enhanced image data from the data preprocessing unit, uses a convolutional neural network to extract features from the images, generates feature vectors, calculates the dimension of the feature vectors for evaluating the number of extracted features, and calculates the mean and standard deviation of the feature vectors for feature normalization, and outputs the normalized feature vectors to the classification and recognition unit; The classification and recognition unit receives the feature vectors from the feature extraction unit, classifies the feature vectors through a decision tree, and calculates the confusion matrix for analyzing the classification results, and outputs the classification results to the visualization output unit; The visualization output unit receives the recognition results from the classification and recognition unit, annotates the recognition results on the original images, calculates the coordinates of the annotation boxes, and outputs the detailed information of the recognized images to the visualization terminal.

[0006] Preferably, the data acquisition unit includes an image recognition module that can automatically identify and classify image types through image processing algorithms for automatic annotation and management of image data.

[0007] Preferably, the data preprocessing unit uses an adaptive denoising algorithm to dynamically adjust the denoising parameters based on the noise characteristics of the images.

[0008] Preferably, the feature extraction unit uses a deep convolutional neural network architecture, including multiple convolutional layers, pooling layers, and fully connected layers, to perform multi-level feature extraction on the images.

[0009] Preferably, the feature extraction unit implements a dynamic adjustment mechanism for feature vectors, which can adaptively increase or decrease the feature dimension according to the complexity of the classification task.

[0010] Preferably, the calculation formula for the dimension of the feature vectors is as follows: In the formula, represents the dimension of the feature vector, represents the feature vector, which is determined by the output layer of the convolutional neural network.

[0011] Preferably, the classification and recognition unit implements an ensemble learning framework, combines multiple base classifiers, including decision trees, random forests, and support vector machines, and performs a voting mechanism to improve the classification accuracy and reduce the bias of a single model.

[0012] Preferably, the classification and recognition unit has the ability of online learning, can receive new training data in real time and dynamically update the classification model to adapt to the change of data distribution.

[0013] Preferably, the calculation formula of the confusion matrix is as follows: In the formula, represents the value at the position in the confusion matrix, represents the number of samples with the true class being and the predicted class being , represents the true class of the th sample, represents the predicted class of the th sample, represents the total number of samples.

[0014] Preferably, the calculation formula of the coordinates of the annotation box is as follows: In the formula, represents the upper left coordinate of the annotation box, represents the lower right coordinate of the annotation box, represents the initial coordinate of the upper left corner, represents the width of the annotation box, represents the height of the annotation box.

[0015] Compared with the prior art, the present invention provides an image recognition system based on artificial intelligence, which has the following beneficial effects: In the data preprocessing stage of the present invention, advanced denoising technology is adopted to significantly improve the image quality, thereby reducing the negative impact of data noise on the classification and recognition accuracy. By calculating the noise level of the denoised image and generating a normalized color histogram, the consistency of the image in the feature extraction process is ensured. The feature extraction unit uses a convolutional neural network to generate feature vectors and systematically evaluates and normalizes them, enabling effective comparison of features between different images, thereby improving the efficiency and accuracy of feature extraction. The classification and recognition unit analyzes the feature vectors through a decision tree algorithm and calculates the confusion matrix, further ensuring the reliability of the classification results. This series of improvements not only enhances the overall performance and reliability of the image recognition system but also provides more accurate image information output in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the system process of the present invention. Specific embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying 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 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.

[0018] In view of the problems of low performance and reliability of the existing image recognition system, an image recognition system based on artificial intelligence is proposed. Please refer to Figure 1 , this system includes a data acquisition unit, a data preprocessing unit, a feature extraction unit, a classification and recognition unit, and a visualization output unit; The data acquisition unit is the core component of the image recognition system. This module contains an image recognition module responsible for batch collecting images from the image library for subsequent processing and analysis. This unit connects to the database or file system storing images through an automated script or program interface, and systematically extracts the required image data. During the acquisition process, in addition to obtaining the pixel information of each image, the data acquisition unit will also record the metadata related to the image for subsequent analysis and management. These metadata include: Shooting time: Record the specific date and time when the image was taken, which helps analyze the timeliness of the image data and its change trend within a specific time period; Shooting location: Record the geographical location where the image was taken, usually stored in the form of longitude and latitude or a specific address. This is crucial for subsequent geographical information analysis and research on application scenarios; Shooting device information: including camera model, lens parameters, shooting settings (such as ISO, shutter speed, aperture, etc.). These information can help researchers understand the source of the image quality and its shooting conditions, and then make corresponding adjustments and optimizations during the feature extraction and classification and recognition processes; Through systematic data acquisition and metadata recording, the data acquisition unit not only improves the efficiency of image processing, but also provides rich background information for subsequent data analysis and applications, ensuring that the entire image recognition system can achieve the best performance in terms of accuracy and effectiveness; The data preprocessing unit plays a crucial role in the image recognition system. It is mainly responsible for cleaning and optimizing all image data from the data acquisition unit to improve the accuracy of subsequent feature extraction and classification recognition. First, this unit adjusts the size of all images to ensure they have a unified resolution. Usually, a target size is set first, and interpolation methods such as bilinear interpolation or cubic interpolation are used to minimize the impact on image quality during the adjustment process; Subsequently, the data preprocessing unit will use various denoising techniques to process the images, including Gaussian filtering, median filtering, or wavelet transform, etc., to effectively remove random noise in the images. This process not only improves the clarity of the images but also can evaluate the effectiveness of denoising by calculating the noise level (such as standard deviation) to ensure that subsequent analysis is based on high-quality data; Next, the preprocessing unit will calculate the color histogram of the images. This process involves counting the color information of each pixel to generate a histogram reflecting the color distribution of the images. For the convenience of subsequent processing and comparison, the generated color histogram data will be normalized to ensure that its numerical range is within a certain interval, thereby eliminating the dimensionality impact between different images; Finally, to further improve the performance of the images in the feature extraction process, the data preprocessing unit will enhance all images. Commonly used enhancement techniques include histogram equalization, contrast stretching, and gamma correction, etc. After these processes, the optimized image data will be transmitted to the feature extraction unit to lay the foundation for subsequent feature learning and classification recognition. Through this series of refined preprocessing steps, the system can extract useful information from the images more accurately and efficiently, thereby improving the performance of the entire image recognition; Among them, the formula for image denoising is as follows: After denoising, the feature extraction algorithm can more accurately identify and learn meaningful patterns, reducing the risk of misrecognition. In the formula, represents the pixel value at position after image denoising, represents the pixel value at position in the original image, represents the number of neighborhoods around the current pixel, and represent the range of the neighborhood, , represents the counting subscript. A clear image helps the model better understand the image content, thereby improving the accuracy of classification or detection. After denoising, the recognition rate can be significantly improved, especially in images taken under complex backgrounds or low-light conditions; The formula for calculating the noise level of the denoised image is as follows: Understanding the noise level of an image helps to adjust other image processing steps (such as feature extraction and classification) to ensure that their processing adapts to the quality of the image. In the formula, represents the noise level of the image, represents the total number of pixels in the image, represents the value of the -th pixel in the denoised image, represents the index sequence. Monitoring the noise level can provide feedback for the image acquisition and processing loop, which helps to continuously improve the system; The formula for calculating the color histogram of an image is as follows: The color histogram can reflect the distribution of each color in the image and provide basic information for subsequent feature extraction and analysis, such as target detection for specific colors. In the formula, represents the color in the image the number of occurrences, represents the position of the pixel in the image, represents the original image, represents the image at the position the color value, represents the indicator function. When is equal to , is 1, otherwise it is 0. By analyzing the color histogram, the color structure of the image can be grasped as a whole, which is convenient for subsequent comparison and classification; The formula for image enhancement is as follows: After the enhancement process, the contrast and brightness of the image can be improved, making it easier to identify the target. In the formula, represents the pixel value of the enhanced image at , , respectively represent the minimum and maximum pixel values of the original image, and represent the target pixel value range, represents the pixel value of the original image at . The enhancement technology can remove the non-uniformity caused by the shooting conditions in the image, enabling the feature extraction algorithm to work more effectively; The feature extraction unit plays a crucial role in the image recognition system. Its main function is to extract meaningful features from the enhanced image data sent by the data preprocessing unit for subsequent classification and recognition. In this process, the feature extraction unit uses a convolutional neural network as the core processing tool, which is widely used because it can automatically learn and extract hierarchical features when processing image data; First, the feature extraction unit will receive the image data that has been denoised and enhanced, and input this data into a pre-trained convolutional neural network. The CNN gradually extracts the spatial features and local patterns of the image through multiple convolutional layers and pooling layers. This process usually includes performing a convolution operation on the image using a filter (convolution kernel) to extract low-level features (such as edges, corners, etc.), and then introducing non-linearity through an activation function (such as ReLU) so that the network can learn more complex features; After several layers of convolution and pooling, the feature extraction unit will generate a series of feature vectors. These feature vectors represent various feature information of the input image and contain important recognition information of the corresponding parts of the image. Next, the system will calculate the dimension of the feature vectors to evaluate the number of extracted features. The dimension of the feature vectors not only determines the input size of the subsequent classifier but also can reflect the complexity and richness of the features extracted by the network; To further improve the usability of the features and the stability of the model, the feature extraction unit will calculate the mean and standard deviation of the feature vectors. This process involves statistical analysis of the generated feature vectors to achieve feature normalization. The feature mean is usually used for centering, while the standard deviation is used for scaling to ensure that the feature data is within the same numerical range. This normalization process can reduce the magnitude difference between features, which helps the training of subsequent classification models (such as support vector machines or deep learning models) and improves the robustness of the model to data changes; Finally, the normalized feature vectors will be output to the classification and recognition unit. These feature vectors are the basis for the model to perform efficient classification and can effectively support the recognition and decision-making tasks of the system in different application scenarios. Through this series of feature extraction and normalization steps, the entire image recognition system can more accurately understand and recognize the content in the image, thereby improving the processing efficiency and recognition accuracy; Among them, the calculation formula for the dimension of the feature vector is as follows: The dimension of the feature vector directly affects the expressive ability of the model. An appropriate dimension selection helps to improve the generalization ability and accuracy of the model. In the formula, represents the dimension of the feature vector, Denote the feature vector, which is determined by the output layer of the convolutional neural network. Reasonably controlling the dimension of the feature vector can reduce the complexity of subsequent calculations, thereby improving the processing efficiency, especially on large-scale image datasets; The formula for calculating the mean of the feature vector is as follows: After calculating the feature mean, centering can be performed to make the model more stable during training and avoid the deviation between features from affecting the learning process. In the formula, denotes the mean of the feature vector, denotes the dimension of the feature vector, denotes the th component of the feature vector, denotes the index subscript. By mean normalization, the influence of noise in the training data on the model can be reduced, thereby alleviating the overfitting phenomenon and improving the generalization ability of the model; The formula for calculating the standard deviation of the feature vector is as follows: The standard deviation can reflect the dispersion degree of the features, thereby helping to understand the stability of the features and identifying unstable features for elimination or adjustment. In the formula, denotes the standard deviation of the feature vector, denotes the th component of the feature vector, denotes the mean of the feature vector, denotes the dimension of the feature vector, denotes the index subscript. Knowing the standard deviation of the features can help adjust the model parameters, making the model more adaptable to different features and improving the accuracy of the model; The classification and recognition unit plays a key role in the image recognition system. It is responsible for receiving the feature vectors processed by the feature extraction unit and classifying the input image based on these features. This unit usually uses the decision tree algorithm as the main classification tool. The decision tree is a commonly used supervised learning model, and its basic principle is to learn the relationship between data features and corresponding categories by creating a model. Specifically, the classification and recognition unit first constructs a decision tree model by using the feature vector as the input. In the training stage, the decision tree uses the information in the feature vector to gradually split the dataset, and each time it selects the feature that can best distinguish the categories for splitting until the preset stopping condition is reached or the classification is complete; After the construction of the decision tree model is completed, the system will predict new samples based on the feature vectors and output the classification results for each sample. To evaluate the performance of the classification model and analyze the quality of the classification results, the classification and recognition unit will calculate the confusion matrix. The confusion matrix is an important tool that can visually display the relationship between the true classes and the predicted classes. It shows the true classes on the rows and the classes predicted by the model on the columns. Therefore, by analyzing the confusion matrix, one can intuitively understand the accuracy, recall rate, and misclassification situations of the model for each class, enabling developers to optimize the model targeted; Among them, the calculation formula of the confusion matrix is as follows: By examining the confusion matrix, one can identify which classes are confused, thus providing a direction for model improvement, such as adjusting the classification boundary or reconsidering the feature extraction method. In the formula, represents the value at position in the confusion matrix, indicating the number of samples with the true class being and the predicted class being . represents the true class of the th sample, represents the predicted class of the th sample, represents the total number of samples. Using the analysis results of the confusion matrix, one can gradually optimize the model training process and select more appropriate hyperparameters, thereby effectively improving the overall performance of the model; The visualization output unit plays an important bridging role in the image recognition system. It is responsible for presenting the recognition results output by the classification and recognition unit to the user in an intuitive manner. This process begins with receiving the classification results and relevant information from the classification and recognition unit. On this basis, the system will use computer vision technology to process the original image. Specifically, by analyzing the recognition results, the coordinates of the annotation box will be automatically calculated to frame the recognized object area; Among them, the calculation formula of the coordinates of the annotation box is as follows: The annotation box can visually display the recognition results, facilitating users or researchers to understand the working effect of the model and facilitating further adjustment and optimization. In the formula, represents the upper left coordinate of the annotation box, represents the lower right coordinate of the annotation box, represents the initial coordinate of the upper left corner, represents the width of the annotation box, Denote the height of the bounding box. Accurate bounding box coordinates not only facilitate the analysis of the current image but also lay the foundation for the comparison between multiple images or the analysis of changes in time-series data; In the process of calculating the bounding box, the system first determines the position and boundaries of each recognized object in the image, which usually involves the conversion of image coordinates, including extracting the corresponding pixel coordinate information from the class label and its confidence value. These coordinates represent the upper-left and lower-right corners of the recognized object, thus forming a rectangular box to highlight these recognition results. To enhance the visualization effect, the bounding box usually adopts a distinct color and has a certain transparency to ensure that the details of the original image are still visible; After completing the calculation and drawing of the bounding box, the system will also update the detailed information of the image, including the recognized object category, confidence score, and other relevant attributes (such as size, position, etc.). These information will be integrated together to facilitate the user to understand the relevant information of each recognized object and help the user make more informed decisions; Finally, these processed images and object information will be output to the visualization terminal. The visualization terminal can be a graphical user interface (GUI), which displays the enhanced image in a graphical way and shows the corresponding text information beside or below it. The user can directly observe the effect of image recognition through this interface, check the recognition accuracy, and quickly understand the working process of the system.

[0019] Through the comprehensive application of the above system, not only the overall performance and reliability of the image recognition system are improved, but also more accurate image information output is provided in the practical application of image recognition.

[0020] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition system based on artificial intelligence, characterized in that: It includes a data acquisition unit, a data preprocessing unit, a feature extraction unit, a classification and recognition unit, and a visualization output unit; The data acquisition unit records metadata of each image, including shooting time, location and shooting equipment information, by batch collecting images in the image library; The data preprocessing unit extracts all images from the data acquisition unit, resizes all images, removes noise from all images using a denoising technique, calculates the noise level of the denoised images, calculates the color histogram of the images, generates normalized color histogram data, and enhances all images before outputting them to the feature extraction unit; The feature extraction unit receives the enhanced image data from the data preprocessing unit, uses a convolutional neural network to extract features from the image, generates a feature vector, calculates the dimension of the feature vector for evaluating the number of extracted features, and calculates the mean value of the feature vector and the standard deviation of the feature vector for feature normalization, and outputs the normalized feature vector to the classification and recognition unit; The classification recognition unit receives the feature vector from the feature extraction unit, classifies the feature vector through a decision tree, and calculates a confusion matrix for analyzing the classification result, and outputs the classification result to the visualization output unit; The visualization output unit receives the recognition result from the classification and recognition unit, marks the recognition result on the original image, calculates the coordinates of the marking box, and then outputs the recognized image detailed information to the visualization terminal.

2. The image recognition system based on artificial intelligence according to claim 1, characterized in that: The data acquisition unit includes an image recognition module, which can automatically recognize and classify image types through image processing algorithms, and is used for automatic labeling and management of image data.

3. The image recognition system based on artificial intelligence according to claim 2, characterized in that: The data preprocessing unit adopts an adaptive denoising algorithm to dynamically adjust denoising parameters based on the noise characteristics of the image.

4. The image recognition system based on artificial intelligence according to claim 3, characterized in that: The feature extraction unit uses a deep convolutional neural network architecture, including multiple convolutional layers, pooling layers and fully connected layers, to perform multi-level feature extraction on the image.

5. The image recognition system based on artificial intelligence according to claim 4, characterized in that: The feature extraction unit implements a dynamic adjustment mechanism of the feature vector, and can adaptively increase or decrease the feature dimension according to the complexity of the classification task.

6. The artificial intelligence-based image recognition system according to claim 5, characterized in that: The calculation formula of the dimension of the feature vector is as follows: In the formula, represents the dimension of the feature vector, Represents the feature vector, which is determined by the output layer of the convolutional neural network.

7. The artificial intelligence-based image recognition system according to claim 6, characterized in that: The classification and recognition unit implements an integrated learning framework, combines multiple base classifiers, including decision trees, random forests, and support vector machines, and performs a voting mechanism to improve classification accuracy and reduce the bias of a single model.

8. The artificial intelligence-based image recognition system according to claim 7, characterized in that: The classification and recognition unit has online learning capability, can receive new training data in real time and dynamically update the classification model to adapt to changes in data distribution.

9. The artificial intelligence-based image recognition system according to claim 8, characterized in that: The calculation formula of the confusion matrix is ​​as follows: In the formula, Represents the position in the confusion matrix The value of indicates that the true category is , the predicted category is The number of samples, Indicates The true category of samples, Indicates The predicted category of samples, Indicates the total number of samples.

10. The image recognition system based on artificial intelligence according to claim 9, characterized in that: The calculation formula of the coordinates of the annotation box is as follows: In the formula, Indicates the coordinates of the upper left corner of the annotation box. Indicates the coordinates of the lower right corner of the annotation box. Indicates the initial coordinates of the upper left corner, Indicates the width of the annotation box. Indicates the height of the callout box.

Images