Distributed artificial intelligence image recognition analysis method and device
Through the distributed artificial intelligence image recognition analysis method, through distributed extraction and feature association, the problems of high computing resource consumption and serious noise impact in image recognition are solved, and efficient and accurate image recognition is achieved.
Patent Information
- Application Number
- CN202510082532.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-20
- Publication Date
- 2025-05-06
AI Technical Summary
The prior art has problems such as high computing resource consumption and serious noise impact in image recognition, especially when images are complex and many features are numerous.
Through the distributed artificial intelligence image recognition analysis method, image data is distributed and extracted, and the image and text transformation correlation is performed. The relationship chain that best fits the key features is obtained through feature cross-correlation, and then denoising is performed during image convolution.
It realizes efficient image recognition, reduces computing resource consumption, reduces noise impact, and improves the accuracy of recognition.
Smart Images
Figure CN119942305A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a distributed artificial intelligence image recognition and analysis method and device. Background Art
[0002] Image recognition refers to the process of understanding, analyzing and classifying information such as targets, scenes, and features in an image using computer technology. Through specific algorithms and models, it enables computers to recognize the content in an image like human eyes, such as identifying people, animals, objects, text, etc. in the image, and determining their category, location, posture and other related information.
[0003] With the development of artificial intelligence technology, it is very common to use convolutional neural networks for intelligent image recognition. It automatically extracts features from data through components such as convolutional layers, pooling layers, and fully connected layers. The convolutional layer extracts local features of the input data through convolution operations, while the pooling layer is used to reduce the dimension of the feature map and reduce computational complexity. The fully connected layer is located at the end of the network and is used to map the extracted features to the output space.
[0004] However, current image recognition has the following problems: 1. There is a lack of literature explaining the use of image data. When the image features are numerous and complex, it not only involves a large number of convolution and pooling operations, but also takes a long time to train and requires high computing resources. 2. When the image is complex, it often carries a large amount of noise that is unrelated to the image elements, which is difficult to avoid during convolutional neural network operations, which not only increases the difficulty of operation, but also affects the accurate recognition of the image. Summary of the invention
[0005] The present invention provides a distributed artificial intelligence image recognition and analysis method and device, which extracts different features of the image after distributing and extracting the image data, associates the features through image-text conversion, and then obtains the relationship chain that best fits all key features through mutual cross-correlation between the features. Then, based on the relationship chain, intelligent recognition after denoising is performed during image convolution to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solution: a distributed artificial intelligence image recognition and analysis method, comprising the following steps: S1: Copy the image A to be recognized to obtain a copy image B with the same elements; S2: Perform convolution operation on image B; S3: Image B enters the convolutional layer of the convolutional neural network to extract local features of image B; S4: classify or predict the data to obtain multiple image features in image B; S5: establishing association relationships for the image features of image B, and performing cross-correlation of the image features; S6: Find the most relevant feature through the correlation relationship of image B; S7: Perform convolution operation on image A, and add the association relationship of image B to the fully connected layer of image A to achieve understanding and judgment of the entire data, finally extract the corresponding associated features of image A, and perform intelligent recognition based on the association relationship.
[0007] Preferably, the specific steps of S3 are: The image B is convolved. The convolution operation includes the convolution kernel set in the convolution layer. The convolution kernel is the parameter of the convolution layer and is a small matrix. By sliding on the input data and performing the convolution operation, the convolution kernel in the convolution layer will slide on the image and perform multiplication and addition operations between elements, capturing local features such as edges, textures, and color distribution, and extracting features from the input. After multiple layers of convolution and pooling, the fully connected layer maps these high-level features to the category space and generates a probability score for each category by using the softmax function.
[0008] Preferably, the specific steps of S4 are: By using the probabilities of different image features, we stratify them and sort the features with high probability and low probability. According to the set probability lower limit, we select the part above the set probability lower limit. Then, we set the priority level according to the selected feature part. The one with high probability has high priority. Then, we get the priority features. After marking, we get the priority image features.
[0009] Preferably, the specific steps of S5 are: The obtained priority image features are converted into text, and then the text interpretation is input into the external artificial intelligence big data model for cross-matching, and then the association is performed after the input data.
[0010] Preferably, the step S5 further comprises: By cross-correlating the features with each other, the corresponding correlation between the two features can be obtained after inputting into the artificial intelligence big data model, and multiple associated words can be obtained each time the cross-correlations are performed.
[0011] Preferably, the step S5 further comprises: The features are also cross-correlated three by three, and after being input into the artificial intelligence big data model, multiple associated words corresponding to the cross-correlations between the three can be obtained.
[0012] Preferably, the specific steps of S6 are: After each of the above associations, the number of association relationships is obtained, and different numbers of associated words are obtained according to the intimacy of the association relationships.
[0013] Preferably, the step S6 further comprises: Extract keywords from all associated words. If there are many associated words, select the top associated words and associate them in the artificial intelligence big data model to find keywords with high matching degree, and then get the relationship chain that best fits all features and associates all key features.
[0014] Preferably, the specific steps of S7 are: The relationship chain of image B is added to the fully connected layer of image A. After being processed by multiple convolutional layers and pooling layers, image A will finally be connected to the fully connected layer. The fully connected layer learns the relationship chain containing key features, identifies according to the association relationship, and screens out non-associated image features. Then the fully connected layer converts the extracted feature map into the final output of the network, and obtains intelligent and accurate recognition of the image that best matches the association relationship.
[0015] The present invention provides a device for the above-mentioned distributed artificial intelligence image recognition and analysis method, including a computer, wherein the computer includes a convolutional neural network model for performing convolution operations on images, and the convolutional neural network model includes links for capturing images to achieve automatic extraction of images; the computer also has an artificial intelligence big data model for identifying associated words, and the artificial intelligence big data model includes links for text capture.
[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. After distributing and extracting the image data, extract different features of the image, associate them through image-text conversion, and then obtain the relationship chain that best fits all key features through cross-correlation between the features. Then, based on the relationship chain, perform intelligent recognition after denoising during image convolution.
[0017] 2. By using a replica image with the same elements as the original image to denoise and find the key features, the processing will not interfere with the original image.
[0018] 3. Through image-to-text conversion, the text interpretation is then input into an external artificial intelligence big data model for cross-matching, which enables efficient and intelligent association.
[0019] 4. By setting the priority of pairwise cross-correlation and three-way cross-correlation, it is easy to find all the feature combination correlation factors that best reflect the theme in the image and eliminate noise.
[0020] 5. According to the probabilities of different image features, the features with high probabilities and low probabilities are stratified and sorted. According to the set probability lower limit, the part above the set probability lower limit is selected, and then the noise is further eliminated based on the selected feature part. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 It is a schematic diagram of the image recognition method flow of the present invention; Figure 2 It is a schematic diagram of the image recognition convolution operation flow of the present invention; Figure 3 It is a schematic diagram of an image flow chart of setting priority of a feature part of the present invention; Figure 4 It is a schematic diagram of the cross-correlation of image features of the present invention; Figure 5 A schematic diagram of a relationship chain flow chart for obtaining key features of the present invention; Figure 6 This is a schematic diagram of intelligent recognition of image A of the present invention; Figure 7 It is a schematic diagram of the structure of the image recognition and analysis device of the present invention. DETAILED DESCRIPTION
[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0023] See also Figure 1 The present invention provides a distributed artificial intelligence image recognition and analysis method, comprising the following steps: S1: Duplicate the image A to be identified to obtain a duplicate image B with the same elements. By first processing the images with the same elements to find the correlation relationship, find the non-correlated factors for image A and add them to the convolutional fully connected layer learning pool of image A. Then, when performing convolution operations on subsequent image A, identify the non-correlated factors, eliminate the interference of non-correlated lines, reduce noise, avoid overfitting, and achieve accurate recognition; S2: Perform a convolution operation on image B and import it into the convolutional neural network. The input layer receives the original image data. The image usually consists of three color channels (red, green, and blue), forming a two-dimensional matrix that represents the intensity value of the pixel; S3: Image B enters the convolutional layer of the convolutional neural network to extract local features of image B. The activation function introduces nonlinearity, which enables the network to learn complex features. Pooling operation is used for dimensionality reduction. Pooling reduces the spatial dimension of the feature map to reduce the amount of calculation. S4: Get multiple image features in image B. After convolution calculation, get an image with prominent edge contour lines. Connect to the fully connected layer based on the contour line image. The fully connected layer integrates the learned features. Then classify or predict the data through a classifier (such as a softmax classifier) to get multiple image features. The features of image B are noisy and contain many unnecessary features. S5: Establishing association relationships between multiple image features of image B. Considering that there are many image features, features with high matching degree are selected. Through multiple cross-correlations, non-correlated factors can be found. S6: Find the most relevant features through the correlation relationship of image B, that is, the combination of all the features that best reflect the theme in the image; S7: Perform convolution operation on image A, and add the association relationship of image B to the fully connected layer of image A. After being processed by multiple convolutional layers and pooling layers, image A will finally be connected to the fully connected layer. The fully connected layer will integrate the learned features to classify or predict the data. Each neuron in the fully connected layer is connected to all neurons in the previous layer. It combines the local features extracted previously into global features, thereby realizing the understanding and judgment of the entire data, and finally extracting the corresponding associated features of image A, and identifying it according to the association relationship.
[0024] See also Figure 2 As a further embodiment, the image B itself contains multiple features, namely feature B1, feature B2, feature B3...feature Bn, where features refer to different objects, such as buildings, cars, animals, plants, etc. in an image. By performing a convolution operation on the image B, the convolution operation includes a convolution kernel set in the convolution layer. The convolution kernel is a parameter of the convolution layer and is a small matrix. By sliding on the input data and performing a convolution operation, the convolution kernel (or filter) in the convolution layer will slide on the image and perform multiplication and addition operations between elements, thereby capturing local features such as edges, textures, and color distributions, and can extract features B1, feature B2, feature B3...feature Bn in the input. The size of the convolution kernel is selected as 3×3 or 5×5. After multiple layers of convolution and pooling, the fully connected layer maps these high-level features to the category space, and generates a probability score for each category by using the softmax function.
[0025] The mathematical expression of the softmax function is softmax(z_i) = e^(z_i) / Σ(j=1 to K) e^(z_j), where z is a real number vector, z_i is the i-th element in the vector, e^(z_i) is the exponent of z_i, and the denominator is the sum of the exponents of all z_j elements. The original output value (that is, the output of the neural network) is converted into a probability distribution. The softmax value of each element represents the "importance" or "probability" of the element relative to all elements in the vector. It is used to find the relationship between each contour line, and then extract buildings, cars, animals, plants, etc.
[0026] See also Figure 3 As a further embodiment, the probabilities of different image features are stratified, and features with high probabilities and features with low probabilities are sorted. According to the set probability lower limit, the part above the set probability lower limit is selected, and then the priority level is set according to the selected feature part, and the priority level is high for the one with high probability, and then the priority features are obtained. After marking, the priority image features Y are obtained, which are image feature Y1, image feature Y2...image feature Yn.
[0027] See also Figure 4 As a further embodiment, the obtained priority image features are converted into images and texts, that is, the image features are given text interpretations, and then the text interpretations are input into an external artificial intelligence big data model for cross-matching. The big data intelligent model can select conventional models on the market, such as ChatGPT, and associate them after inputting the data.
[0028] See also Figure 4 As a further embodiment, the features are then cross-correlated with each other, that is, Y1 is cross-correlated with Y2, Y1 is cross-correlated with Y3, Y2 is cross-correlated with Y3, ... Yn-1 is cross-correlated with Yn, and after inputting into the ChatGPT model, the corresponding association relationship between the two can be obtained, such as the association between car and road, and the association words such as car parked on the road or driving on the road can be obtained. Multiple association words can be obtained each time the cross is crossed.
[0029] See also Figure 4 As a further embodiment, when there are many features, cross-correlations are performed between three features, such as Y1, Y2, and Y3, Y2, Y3, and 4, Y1, Y3, and 4, ... Y1, Yn-1, and Yn are cross-correlated to obtain associations between three features. For example, after inputting coconut trees, beaches, and surfboards into the ChatGPT model, multiple associated words such as surfing on the beach in summer and traveling to the beach are obtained, and then multiple associated words are obtained after the associations between three features.
[0030] See also Figure 5 As a further embodiment, after each association, the number of associations will be obtained, that is, different numbers of associated words will be obtained according to the intimacy of the associations. For example, the associations of high-speed rail, badminton, and straws are low, and thus fewer associated words will be obtained in ChatGPT, while the associations of shopping malls, mobile phones, and toys are more, such as buying toys in a shopping mall through a mobile phone, taking mobile phone toys to a shopping mall to take pictures, and so on.
[0031] See also Figure 5 As a further embodiment, all associated words are used to extract keywords. If there are many associated words during extraction, the top associated words are selected and then associated in ChatGPT to find keywords with high matching degree, and then the most suitable relationship chain is obtained by associating all key features, and the relationship chain only contains key features.
[0032] See also Figure 6 As a further embodiment, the relationship chain of image B is added to the fully connected layer of image A. After being processed by multiple convolutional layers and pooling layers, image A is finally connected to the fully connected layer. The fully connected layer learns the relationship chain containing key features, identifies according to the association relationship, and screens out non-associated image features. Then, the fully connected layer converts the extracted feature map into the final output of the network, and obtains intelligent and accurate recognition of the image that best conforms to the association relationship.
[0033] See also Figure 7 The present invention provides a device applied to the above-mentioned distributed artificial intelligence image recognition and analysis method, including a computer, wherein the computer includes a convolutional neural network model for performing convolution operations on images, and the convolutional neural network model includes links for crawling images to achieve automatic extraction of images; the computer also has an artificial intelligence big data model for identifying associated words, such as a ChatGPT model, and the artificial intelligence big data model includes links for crawling text.
[0034] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A distributed artificial intelligence image recognition and analysis method, characterized in that: The steps include: S1: Copy the image A to be recognized to obtain a copy image B with the same elements; S2: Perform convolution operation on image B; S3: Image B enters the convolutional layer of the convolutional neural network to extract local features of image B; S4: classify or predict the data to obtain multiple image features in image B; S5: establishing association relationships for the image features of image B, and performing cross-correlation of the image features; S6: Find the most relevant feature through the correlation relationship of image B; S7: Perform convolution operation on image A, and add the association relationship of image B to the fully connected layer of image A to achieve understanding and judgment of the entire data, finally extract the corresponding associated features of image A, and perform intelligent recognition based on the association relationship.
2. A distributed artificial intelligence image recognition and analysis method according to claim 1, characterized in that: The specific steps of S3 are: The image B is convolved. The convolution operation includes the convolution kernel set in the convolution layer. The convolution kernel is the parameter of the convolution layer and is a small matrix. By sliding on the input data and performing the convolution operation, the convolution kernel in the convolution layer will slide on the image and perform multiplication and addition operations between elements, capturing local features such as edges, textures, and color distribution, and extracting features from the input. After multiple layers of convolution and pooling, the fully connected layer maps these high-level features to the category space and generates a probability score for each category by using the softmax function.
3. A distributed artificial intelligence image recognition and analysis method according to claim 2, characterized in that: The specific steps of S4 are: By using the probabilities of different image features, we stratify them and sort the features with high probability and low probability. According to the set probability lower limit, we select the part above the set probability lower limit. Then, we set the priority level according to the selected feature part. The one with high probability has high priority. Then, we get the priority features. After marking, we get the priority image features.
4. A distributed artificial intelligence image recognition and analysis method according to claim 3, characterized in that: The specific steps of S5 are: The obtained priority image features are converted into text, and then the text interpretation is input into the external artificial intelligence big data model for cross-matching, and then the association is performed after the input data.
5. A distributed artificial intelligence image recognition and analysis method according to claim 4, characterized in that: The steps of S5 also include: By cross-correlating the features with each other, the corresponding correlation between the two features can be obtained after inputting into the artificial intelligence big data model, and multiple associated words can be obtained each time the cross-correlations are performed.
6. A distributed artificial intelligence image recognition and analysis method according to claim 5, characterized in that: The steps of S5 also include: The features are also cross-correlated three by three, and after being input into the artificial intelligence big data model, multiple associated words corresponding to the cross-correlations between the three can be obtained.
7. A distributed artificial intelligence image recognition and analysis method according to claim 6, characterized in that: The specific steps of S6 are: After each of the above associations, the number of association relationships is obtained, and different numbers of associated words are obtained according to the intimacy of the association relationships.
8. A distributed artificial intelligence image recognition and analysis method according to claim 7, characterized in that: The steps of S6 also include: Extract keywords from all associated words. If there are many associated words, select the top associated words and associate them in the artificial intelligence big data model to find keywords with high matching degree, and then get the relationship chain that best fits all features and associates all key features.
9. A distributed artificial intelligence image recognition and analysis method according to claim 8, characterized in that: The specific steps of S7 are: The relationship chain of image B is added to the fully connected layer of image A. After being processed by multiple convolutional layers and pooling layers, image A will finally be connected to the fully connected layer. The fully connected layer learns the relationship chain containing key features, identifies according to the association relationship, and screens out non-associated image features. Then the fully connected layer converts the extracted feature map into the final output of the network, and obtains intelligent and accurate recognition of the image that best matches the association relationship.
10. A device applied to the distributed artificial intelligence image recognition and analysis method according to any one of claims 1 to 9, characterized in that: The invention comprises a computer, wherein the computer comprises a convolutional neural network model for performing convolution operations on images, wherein the convolutional neural network model comprises a link for capturing images, thereby realizing automatic extraction of images; the computer also comprises an artificial intelligence big data model for identifying associated words, wherein the artificial intelligence big data model comprises a link for capturing texts.