Computer image recognition method
Through image preprocessing and convolutional neural network feature extraction, combined with support vector machine classifier, the accuracy problem of image recognition in the prior art in complex environments is solved, and more efficient image recognition effect is achieved.
Patent Information
- Application Number
- CN202411825326.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-12
- Publication Date
- 2025-04-01
AI Technical Summary
In complex environments, the existing computer image recognition methods are not accurate enough to extract image features and are easily disturbed by factors such as lighting, resulting in low recognition accuracy.
Grayscale and filtering processing in image preprocessing are used, feature extraction is combined with convolutional neural network, and classification recognition is performed using support vector machine classifier.
It improves the quality of the image, reduces the interference of external factors on image recognition, improves the accuracy of the recognition, and has high computing efficiency when processing large-scale image data, meeting real-time requirements.
Smart Images
Figure CN119741547A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a computer image recognition method. Background Art
[0002] Image recognition refers to the technology of using computers to process, analyze and understand the image to be recognized in order to detect and recognize target objects of various different patterns. At present, with the development of image recognition technology, image recognition technology has been widely used in various fields.
[0003] In complex environments, the existing computer image recognition methods are not accurate enough in extracting image features and are easily affected by factors such as lighting, resulting in low recognition accuracy. This can easily cause the image to be inaccurately recognized during the recognition process. Therefore, it is necessary to design a computer image recognition method. Summary of the invention
[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a computer image recognition method.
[0005] In order to achieve the above object, the present invention adopts the following technical solution: a computer image recognition method, comprising the following steps:
[0006] S1: The image acquisition device acquires the image and inputs the image into the computer system in digital form;
[0007] S2: performing image preprocessing on the acquired image;
[0008] S3: extract and optimize features of the processed image;
[0009] S4: Input the extracted features into the support vector machine classifier for classification and recognition.
[0010] The image preprocessing in step S2 includes grayscale processing and filtering processing. The grayscale processing is to convert the color image into a grayscale image, and its formula is:
[0011] Gray=0.299×R+0.587×G+0.114×B
[0012] Where R, G, B are the values of the red, green, and blue channels of the color image, respectively, and Gray is the grayscale value;
[0013] The filtering process is to filter the grayscale image using Gaussian filtering, and the formula is:
[0014]
[0015] Where (x, y) is the image pixel coordinate and σ is the standard deviation.
[0016] The feature extraction in step S3 uses a convolutional neural network for feature extraction, and the convolutional neural network consists of a convolutional layer, a pooling layer and a fully connected layer.
[0017] The convolution layer is the core part of the convolutional neural network. The convolution layer performs convolution operations by sliding the convolution kernel on the image; the pooling layer is used to reduce the amount of data while retaining important feature information. The pooling layer adopts an adaptive pooling method; the fully connected layer integrates the previously extracted features and outputs the final classification result. Each neuron in the fully connected layer is connected to all neurons in the previous layer and calculated through a weight matrix.
[0018] The adaptive pooling method dynamically adjusts the size of the pooling area according to the feature distribution of the image. The formula is:
[0019] P = f(A)
[0020] Where P is the feature map after pooling, A is the input feature map, and f is the adaptive function.
[0021] The weight matrix is usually calculated using the minimum mean square error method, and its calculation formula is:
[0022] W=(H T ×H+λI) -1 ×H T ×Y
[0023] Among them, W is the weight matrix, H is the input matrix, Y is the output matrix, λ is the regularization parameter, and I is the identity matrix.
[0024] The optimization in step S3 is to use the model pre-trained on a large-scale data set as a feature extractor and then fine-tune the specific task; through transfer learning, the generalization ability of the pre-trained model on big data is utilized to improve the performance of specific tasks and reduce the required amount of training data and training time.
[0025] The classification recognition in step S4 adopts a polynomial kernel function when training the support vector machine classifier, and its formula is: K(x,y)=(x*y+c) d , where x and y are input feature vectors, c is a constant, and d is the degree of the polynomial.
[0026] The present invention has the following beneficial effects:
[0027] 1. Compared with the prior art, the present invention improves the quality of the image and reduces the interference of external factors on image recognition through grayscale and filtering processing in image preprocessing.
[0028] 2. Compared with the prior art, the present invention adopts a convolutional neural network structure when extracting features from images, which can extract more comprehensive and effective image features and improve the recognition accuracy. The adaptive pooling method in the convolutional neural network structure and the support vector machine classifier in classification recognition further enhance the performance of image recognition. At the same time, when processing large-scale image data, the present invention has higher computational efficiency and can meet application scenarios with higher real-time requirements. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION
[0030] Reference Figure 1 The present invention provides a computer image recognition method, comprising the following steps:
[0031] S1: The image acquisition device acquires the image and inputs the image into the computer system in digital form;
[0032] S2: performing image preprocessing on the acquired image;
[0033] S3: extract and optimize features of the processed image;
[0034] S4: Input the extracted features into the support vector machine classifier for classification and recognition.
[0035] Specifically, when applied, a camera is used to acquire images, and the images captured by the camera are input into a computer system in a digital form. The captured images are then preprocessed. During preprocessing, the computer's image processing library is used to grayscale and filter the images. Feature extraction and optimization are performed on the processed images. Finally, the support vector machine classifier is used for classification and recognition. The recognized image features are compared with the sample image features to determine whether the image is a target image.
[0036] The image preprocessing in step S2 includes grayscale processing and filtering processing. Grayscale processing is to convert a color image into a grayscale image, and its formula is:
[0037] Gray=0.299×R+0.587×G+0.114×B.
[0038] Where R, G, B are the values of the red, green, and blue channels of the color image, respectively, and Gray is the grayscale value;
[0039] Filtering is to filter the grayscale image using Gaussian filtering, and the formula is:
[0040]
[0041] Where (x, y) is the image pixel coordinate and σ is the standard deviation.
[0042] When performing image preprocessing, the image is first grayscaled. This step can reduce the amount of image data and reduce the impact of factors such as lighting on color. The grayscale processed image is then filtered. This step can effectively remove noise from the image.
[0043] In step S3, feature extraction is performed using a convolutional neural network, which consists of a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is the core of the convolutional neural network. The convolutional layer performs convolution operations by sliding the convolution kernel on the image; the pooling layer is used to reduce the amount of data while retaining important feature information. The pooling layer uses an adaptive pooling method; the fully connected layer integrates the previously extracted features and outputs the final classification result. Each neuron in the fully connected layer is connected to all neurons in the previous layer and is calculated through a weight matrix.
[0044] Specifically, the convolution layer is the core part of the convolutional neural network. It performs convolution operations by sliding the convolution kernel (also called filter) on the image. For example, a convolution kernel convolves an image area, moving one pixel each time, calculating the dot product between the convolution kernel and the corresponding image area, and obtaining a new feature map. This process can extract local features in the image, such as texture, shape, etc.
[0045] Pooling layer: The pooling layer is mainly used to reduce the amount of data while retaining important feature information. Common pooling methods include adaptive pooling methods. The adaptive pooling method dynamically adjusts the size of the pooling area according to the feature distribution of the image. The formula is: P=f(A), where P is the feature map after pooling, A is the input feature map, and f is the adaptive function. In terms of adaptive learning, the adaptive pooling method can effectively capture and represent the discriminative information in the input data by learning the optimal pooling strategy during the training process. In terms of improving classification performance, experimental results show that it can achieve good performance on three different data sets, with higher accuracy than traditional methods.
[0046] Fully connected layer: The fully connected layer integrates the previously extracted features and outputs the final classification result. Each of its neurons is connected to all neurons in the previous layer and is calculated through the weight matrix. The weight matrix is usually calculated using the minimum mean square error method, and its calculation formula is: W=(H T ×H+λI) -1 ×H T×Y, where W is the weight matrix, H is the input matrix, Y is the output matrix, λ is the regularization parameter, and I is the unit matrix. The weight matrix performs weighted sum fusion on the feature vectors of different modalities. The weight matrix is dynamically updated according to the contribution of different modal features to the recognition results in previous training and the complexity of the current input image. The weight update rule is calculated based on the softmax function combined with feature importance evaluation indicators (such as feature discrimination, relevance, etc.) to ensure that key modal features are highlighted during the fusion process and improve the effectiveness of comprehensive feature expression.
[0047] The optimization in step S3 is to use the model pre-trained on a large-scale dataset as a feature extractor and then fine-tune it for a specific task; through transfer learning, the generalization ability of the pre-trained model on big data is utilized to improve the performance of specific tasks and reduce the amount of training data and training time required.
[0048] The classification recognition in step S4 uses a polynomial kernel function when training the support vector machine classifier, and its formula is: K(x,y)=(x*y+c) d , where x and y are input feature vectors, c is a constant, and d is the polynomial degree. When used, the extracted feature vector is input into the trained support vector machine classifier, the polynomial kernel function is used to calculate the image features, and the image category is determined according to the output of the classifier.
[0049] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A computer image recognition method, characterized in that: The following steps are included: S1: The image acquisition device acquires the image and inputs the image into the computer system in digital form; S2: performing image preprocessing on the acquired image; S3: extract and optimize features of the processed image; S4: Input the extracted features into the support vector machine classifier for classification and recognition.
2. A computer image recognition method according to claim 1, characterized in that: The image preprocessing in step S2 includes grayscale processing and filtering processing. The grayscale processing is to convert the color image into a grayscale image, and its formula is: Gray=0.299×R+0.587×G+0.114×B Where R, G, B are the values of the red, green, and blue channels of the color image, respectively, and Gray is the grayscale value; The filtering process is to filter the grayscale image using Gaussian filtering, and the formula is: ， Where (x, y) is the image pixel coordinate and σ is the standard deviation.
3. A computer image recognition method according to claim 1, characterized in that: The feature extraction in step S3 uses a convolutional neural network for feature extraction, and the convolutional neural network consists of a convolutional layer, a pooling layer and a fully connected layer.
4. A computer image recognition method according to claim 3, characterized in that: The convolution layer is the core part of the convolutional neural network. The convolution layer performs convolution operations by sliding the convolution kernel on the image; the pooling layer is used to reduce the amount of data while retaining important feature information. The pooling layer adopts an adaptive pooling method; the fully connected layer integrates the previously extracted features and outputs the final classification result. Each neuron in the fully connected layer is connected to all neurons in the previous layer and calculated through a weight matrix.
5. A computer image recognition method according to claim 4, characterized in that: The adaptive pooling method dynamically adjusts the size of the pooling area according to the feature distribution of the image. The formula is: P = f(A) Where P is the feature map after pooling, A is the input feature map, and f is the adaptive function.
6. A computer image recognition method according to claim 1, characterized in that: The weight matrix is usually calculated using the minimum mean square error method, and its calculation formula is: W=(H T ×H+λI) -1 ×H T ×Y Among them, W is the weight matrix, H is the input matrix, Y is the output matrix, λ is the regularization parameter, and I is the identity matrix.
7. A computer image recognition method according to claim 1, characterized in that: The optimization in step S3 is to use the model pre-trained on a large-scale data set as a feature extractor and then fine-tune the specific task; through transfer learning, the generalization ability of the pre-trained model on big data is utilized to improve the performance of specific tasks and reduce the required amount of training data and training time.
8. A computer image recognition method according to claim 1, characterized in that: The classification recognition in step S4 adopts a polynomial kernel function when training the support vector machine classifier, and its formula is: K(x,y)=(x*y+c) d , where x and y are input feature vectors, c is a constant, and d is the degree of the polynomial.