A multi-division fusion image feature extraction method and system

Abstract

The application relates to the technical field of image feature extraction, and provides a multi-segmentation fusion image feature extraction method and system, which comprises the following steps: identifying a target in a first to-be-processed image to obtain n targets; performing gray scale processing to output a first gray scale image; obtaining a gray scale pixel set of the n targets and a residual pixel set in the first gray scale image; inputting the gray scale pixel set and the residual pixel set into a gray scale conversion model respectively to perform gray scale conversion and output n groups of converted gray scale images; outputting n segmentation images according to the n groups of gray scale images, performing image feature extraction, and outputting n image features; and performing image fusion processing to output the image features of the first to-be-processed image. The technical problem that the feature extraction difficulty is high and the feature extraction quality is low when the image feature difference is small can be solved, and the efficiency and quality of image feature extraction can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image feature extraction, in particular to a multi-segmentation fusion image feature extraction method and system. BACKGROUND

[0002] Image feature extraction is an important basic work in computer vision, which provides a key data basis for subsequent image recognition, target detection, image classification and other tasks. Different tasks and application scenarios may require different feature extraction methods. When the feature difference in the image is small, the existing image feature extraction method is difficult to distinguish the extracted features in the image, resulting in high difficulty in image feature extraction and poor extraction effect.

[0003] In summary, the prior art has the technical problem of high difficulty and low quality of feature extraction when the feature difference in the image is small. SUMMARY

[0004] Therefore, it is necessary to provide a multi-segmentation fusion image feature extraction method and system to solve the above technical problems.

[0005] A multi-segmentation fusion image feature extraction method, comprising: when a first to-be-processed image is obtained, identifying a target in the first to-be-processed image to obtain n targets, wherein n is a positive integer greater than or equal to 2; performing gray scale processing on the first to-be-processed image to output a first gray scale image; obtaining n sets of gray scale pixel sets of the n targets in the first gray scale image and n sets of remaining pixel sets in the first gray scale image, wherein the n sets of gray scale pixel sets correspond to the n sets of remaining pixel sets; inputting each set of gray scale pixel set and each set of remaining pixel set into a gray scale conversion model for gray scale conversion to output n sets of converted gray scale images; outputting n segmentation images according to the n sets of gray scale images, and performing image feature extraction on the n segmentation images respectively to output n image features; and outputting an image feature of the first to-be-processed image by performing image fusion processing on the n image features.

[0006] In one embodiment, the method further comprises: identifying the n targets in the first to-be-processed image to obtain multiple types of targets; setting a gray scale regulation channel in the gray scale conversion model by using the multiple types of targets, so that the number of channels of the gray scale regulation channel is the same as the number of target types, wherein the gray scale regulation coefficients in each channel are different, and when a target of the same type is input into the gray scale conversion model, the corresponding channel is matched for gray scale regulation.

[0007] In an embodiment, further comprising: obtaining n sets of gray data and n sets of residual gray data according to the n sets of gray pixel sets and the n sets of residual pixel sets; taking the correspondence between the n sets of gray data and the n sets of residual gray data as a training data set, and training based on the training data set with a target function, and outputting the converted n sets of gray images when the response condition of the target function is met; wherein the target function is a function established to increase the difference between the n sets of gray data and the n sets of residual gray data.

[0008] In an embodiment, further comprising: obtaining the converted n sets of gray images, wherein the n sets of gray images are images after the first to-be-processed image is subjected to gray conversion; performing edge recognition on the n sets of gray images through an edge algorithm, and outputting n pre-segmented images; performing edge smoothing processing on the n pre-segmented images, and outputting the n segmented images.

[0009] In an embodiment, further comprising: establishing a connection relationship for each target corresponding to the n image features, and generating target correlation; performing information entropy calculation on the target correlation, and configuring the weight of the information entropy calculation result when feature fusion processing is performed; performing image fusion processing on the n image features again according to the configured weight, and outputting the image features of the first to-be-processed image.

[0010] In an embodiment, further comprising: judging the feature coincidence degree between the n image features, and obtaining an identified image feature when the feature coincidence degree is greater than a preset feature coincidence degree; performing multi-scale transformation on the identified image feature, and outputting a high-dimensional mapping feature; performing image fusion processing on the high-dimensional mapping feature as the corresponding feature of the identified image feature.

[0011] A multi-segmentation fusion image feature extraction system, comprising:

[0012] A target recognition module, configured to recognize a target in a first to-be-processed image after the first to-be-processed image is obtained, and obtain n targets, wherein n is a positive integer greater than or equal to 2;

[0013] A gray processing module, configured to perform gray processing on the first to-be-processed image, and output a first gray image;

[0014] A pixel set obtaining module, configured to obtain n sets of gray pixel sets of the n targets in the first gray image, and n sets of residual pixel sets in the first gray image, wherein the n sets of gray pixel sets correspond to the n sets of residual pixel sets;

[0015] a gray scale conversion module, configured to input a set of gray scale pixel sets and a set of remaining pixel sets into a gray scale conversion model respectively for gray scale conversion, and output n sets of converted gray scale images;

[0016] an image feature extraction module, configured to output n segmentation images according to the n sets of gray scale images, and extract image features of the n segmentation images respectively, and output n image features;

[0017] an image feature output module, configured to output image features of the first image to be processed by performing image fusion processing on the n image features.

[0018] The multi-segmentation fusion image feature extraction method and system can solve the technical problems of high difficulty and low quality of feature extraction when the image feature difference is small. First, after obtaining the first image to be processed, the target in the first image to be processed is identified, and n targets are obtained, where n is a positive integer greater than or equal to 2. The first image to be processed is subjected to gray scale processing, and a first gray scale image is output. N sets of gray scale pixel sets of the n targets in the first gray scale image and n sets of remaining pixel sets in the first gray scale image are obtained, where the n sets of gray scale pixel sets correspond to the n sets of remaining pixel sets. A set of gray scale pixel sets and a set of remaining pixel sets are input into a gray scale conversion model respectively for gray scale conversion, and n sets of converted gray scale images are output. According to the n sets of gray scale images, n segmentation images are output, and image features of the n segmentation images are extracted respectively, and n image features are output. The image features of the first image to be processed are output by performing image fusion processing on the n image features. The above method can improve the efficiency and quality of image feature extraction.

[0019] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of a multi-segmentation fusion image feature extraction method is provided for the present application;

[0021] Figure 2 A flowchart of outputting n segmentation images in a multi-segmentation fusion image feature extraction method is provided for the present application;

[0022] Figure 3A flowchart of an image feature extraction method for outputting image features of a first to-be-processed image is provided in the present application.

[0023] Figure 4 A structural diagram of an image feature extraction system for multi-segmentation fusion is provided in the present application.

[0024] Legend: target recognition module 1, grayscale processing module 2, pixel set acquisition module 3, grayscale conversion module 4, image feature extraction module 5, image feature output module 6. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0026] As shown in Figure 1 The present application provides an image feature extraction method for multi-segmentation fusion, comprising:

[0027] After obtaining the first to-be-processed image, the target in the first to-be-processed image is identified, and n targets are obtained, wherein n is a positive integer greater than or equal to 2;

[0028] Image feature extraction is a key step in the field of image processing and computer vision, which is used to convert the information in the image into numerical form that can be used for analysis and recognition. The goal of image feature extraction is to find representative information in the image, which can be used to describe the appearance, texture, shape, edge and other features of the image.

[0029] Multi-Segmentation Fusion is a technique in image processing and computer vision, which is used to fuse information from multiple image segmentation algorithms or multiple segmentation results to get more accurate and comprehensive segmentation results. This technique can improve the accuracy and robustness of image segmentation, especially suitable for complex scenes and difficult sample segmentation tasks. The method provided in the present application is used to extract image features through multi-segmentation fusion, achieving the purpose of improving the efficiency and quality of image feature extraction.

[0030] First, a first to-be-processed image is obtained, the first to-be-processed image being an image to be subjected to image feature extraction, and then a target in the first to-be-processed image is identified, wherein the target refers to an image feature to be extracted in the first to-be-processed image. The target identification method can be implemented by means of feature comparison, feature detection, manual identification, etc., or a target identification model can be constructed based on a neural network to identify the target, thereby obtaining n targets, wherein n is a positive integer greater than or equal to 2. The n targets obtained by identifying the target provide support for image segmentation in the next step.

[0031] The first to-be-processed image is subjected to grayscale processing, and a first grayscale image is output.

[0032] The first to-be-processed image is subjected to grayscale processing, wherein image grayscale processing is an image processing technique used to convert a color image into a grayscale image. A grayscale image is an image that only contains grayscale values and has no color information, and is usually used to simplify image processing and analysis tasks. In a grayscale image, each pixel has only one grayscale value representing the brightness level of the pixel, and the value range is usually 0 (black) to 255 (white). Common image grayscale processing methods include average method, weighted average method, maximum value method, and minimum value method, etc., and a person skilled in the art can select an appropriate image grayscale processing method to obtain the first grayscale image.

[0033] n sets of grayscale pixel sets of the n targets in the first grayscale image and n sets of remaining pixel sets in the first grayscale image are obtained, wherein the n sets of grayscale pixel sets correspond to the n sets of remaining pixel sets.

[0034] According to the n targets in the first grayscale image, grayscale pixel extraction is performed to obtain n sets of grayscale pixel sets of the n targets in the first grayscale image, wherein the n sets of grayscale pixel sets refer to grayscale pixel sets of the n targets. And n sets of remaining pixel sets in the first grayscale image are obtained, the remaining pixels referring to grayscale pixels of the background corresponding to the target in the first grayscale image, wherein the n sets of grayscale pixel sets and the n sets of remaining pixel sets have a corresponding relationship.

[0035] By obtaining the n sets of remaining pixel sets, the grayscale values of the background are changed to highlight each feature, so that image feature extraction becomes easier, thereby improving the accuracy and efficiency of image feature extraction.

[0036] A set of grayscale pixel sets and a set of remaining pixel sets are respectively input into a grayscale conversion model for grayscale conversion, and n sets of converted grayscale images are output.

[0037] In one embodiment, it further comprises:

[0038] n targets in the first image to be processed are subjected to type recognition to obtain multiple types of targets;

[0039] The multiple types of targets are used to set the gray regulation channels in the gray conversion model, so that the number of the gray regulation channels is the same as the number of target types, wherein the gray regulation coefficients in each channel are different, and when a target of the same type is input into the gray conversion model, the corresponding channel is matched for gray regulation.

[0040] A gray conversion model is constructed, wherein the gray conversion model comprises multiple gray regulation channels, and the gray regulation coefficients in each gray regulation channel are different, the gray regulation coefficient being a regulation weight for gray processing.

[0041] First, n targets in the first image to be processed are subjected to type recognition, which means that the n targets are classified, which can be achieved by dividing the targets into different categories or types according to their appearance, shape, texture and other characteristics, for example, the targets can be divided into types such as people, plants, animals and buildings, to obtain multiple types of targets. Then, the multiple types of targets are used to set the gray regulation channels in the gray conversion model, wherein the number of the gray regulation channels is the same as the number of target types, and the gray regulation coefficients in each channel are different. For example, when the target is a building, the features are more obvious when the gray value is large, so the gray regulation coefficient in the building channel can be increased, which can be set by a person skilled in the art based on actual conditions. When a target of the same type is input into the gray conversion model, the corresponding channel is matched for gray regulation. By constructing the gray conversion model, the gray conversion can be adapted according to the target type, thereby improving the quality of image gray conversion and reducing the difficulty of image feature extraction.

[0042] A set of gray pixel sets and a set of remaining pixel sets are input into a gray conversion model for gray conversion, wherein the set of gray pixel sets is any one of n sets of gray pixel sets, and the set of remaining pixel sets and the set of gray pixel sets have a corresponding relationship. First, the set of gray pixel sets is matched with a suitable gray regulation channel according to the type of the set of gray pixel sets, and then the set of gray pixel sets and the set of remaining pixel sets are subjected to gray conversion according to the gray regulation coefficients in the gray regulation channel, to output n sets of converted gray images.

[0043] In one embodiment, the present application further comprises:

[0044] According to the n sets of gray pixel sets and the n sets of remaining pixel sets, n sets of gray data and n sets of remaining gray data are obtained;

[0045] According to the correspondence between the n sets of gray data and the n sets of residual gray data, a training data set is obtained, and the n sets of gray data are converted based on the training data set and a target function. When the response condition of the target function is met, the converted n sets of gray images are output.

[0046] The target function is a function established to increase the difference between the n sets of gray data and the n sets of residual gray data.

[0047] According to the n sets of gray pixel sets and the n sets of residual pixel sets, n sets of gray data and n sets of residual gray data are obtained. Then, each set of gray data and the corresponding residual gray data is taken as a set of training data according to the correspondence between the n sets of gray data and the n sets of residual gray data, and n sets of training data are obtained.

[0048] The target function, also known as the loss function, is an important mathematical function in machine learning, optimization and optimization problems. It is used to measure the difference or error between the model prediction result and the true label (or target), and is the core part of the optimization problem. The target function is a function established to increase the difference between the n sets of gray data and the n sets of residual gray data, that is, the greater the difference between the gray data and the residual gray data, the better the fitness of the target function. According to the n sets of training data, the target function is trained, and the target function response condition is set. The target function response condition can be set by the person skilled in the art according to the actual situation. The fitness of the target function is judged according to the target function response condition. When the response condition of the target function is met, the converted n sets of gray images are output. By constructing the target function, the quality of image gray conversion can be improved, thereby improving the efficiency of image feature extraction.

[0049] According to the n sets of gray images, n sets of segmentation images are output, and image feature extraction is performed on the n sets of segmentation images respectively, and n sets of image features are output.

[0050] As shown in Figure 2 in one embodiment, comprising:

[0051] The n sets of converted gray images are obtained, wherein the n sets of gray images are the images obtained by converting the first image to be processed;

[0052] The n sets of gray images are edge-identified by an edge algorithm, and n sets of pre-segmentation images are output.

[0053] The n sets of pre-segmentation images are edge-smoothed, and the n sets of segmentation images are output.

[0054] n groups of converted gray scale images are acquired, wherein the n groups of gray scale images are images after the first to-be-processed image is converted into gray scale. Then, edge recognition is performed on the n groups of gray scale images by using an edge algorithm, which is a commonly used image processing algorithm in computer vision and is used for detecting edges or contours in an image. An image edge generally refers to a position where a pixel value changes sharply in an image, and the position represents a boundary of an object or a boundary between objects. n pre-segmented images are output according to the edge recognition results of the n groups of gray scale images.

[0055] Edge smoothing processing is performed on the n pre-segmented images, and the edge smoothing processing reduces or eliminates noise or discontinuity of edges in the images, so that the edges in the images are smoother and more continuous. The edge smoothing processing method includes median filtering, Gaussian filtering, and bilateral filtering, and an appropriate edge smoothing algorithm can be selected according to actual requirements. By performing the edge smoothing processing on the n pre-segmented images, the accuracy of image segmentation can be improved, and n segmented images are obtained.

[0056] Then, image feature extraction is performed on the n segmented images respectively, and the image feature extraction method is the same as described above, and will not be described here, and n image features are obtained.

[0057] Image fusion processing is performed on the n image features, and an image feature of the first to-be-processed image is output.

[0058] As shown in FIG. 1, Figure 3 In one embodiment, the method further includes:

[0059] A connection relationship is established for each target corresponding to the n image features, and target correlation is generated.

[0060] Information entropy is calculated based on the target correlation, and a weight of a feature fusion processing result is configured based on the information entropy calculation result.

[0061] Image fusion processing is performed on the n image features again based on the configured weight, and an image feature of the first to-be-processed image is output.

[0062] In one embodiment, the method further includes:

[0063] A feature coincidence degree between the n image features is determined, and when the feature coincidence degree is greater than a preset feature coincidence degree, an identification image feature is acquired.

[0064] The identification image feature is subjected to multi-scale transformation, and a high-dimensional mapping feature is output.

[0065] The high-dimensional mapping feature is used as a corresponding feature of the identification image feature for image fusion processing.

[0066] The n image features are subjected to image fusion processing. First, a connection relationship is established for each target corresponding to the n image features. The connection relationship is established by marking each target according to the position coordinates of each target in the first image to be processed, and generating target correlation according to the n image features.

[0067] Then, information entropy is calculated according to the target correlation. The information entropy calculation is used to measure the uncertainty or information amount of a random variable. In the calculation of information entropy, it is assumed that the random variable X has n possible values, which are x1, x2,..., xn, and the probability of each value xi occurring is P(xi). The calculation formula of information entropy is: H(X) = -∑(P(xi) * log2(P(xi))); where H(X) represents the information entropy of the random variable X, ∑ represents the summation of all values xi, and log2 represents the logarithm operation with base 2. The information entropy calculation result is obtained, and the weight of the n target feature fusion processing is configured according to the information entropy calculation result. The greater the information entropy calculation result, the greater the feature fusion processing weight of the corresponding target. The weight can be set by the existing coefficient of variation method, which is a commonly used weight setting method by those skilled in the art and will not be described here.

[0068] According to the configured weight, the n image features are subjected to image fusion processing. First, a preset feature coincidence degree is obtained, which can be set by those skilled in the art according to actual conditions. The feature coincidence degree between the n image features is determined according to the preset feature coincidence degree. When the feature coincidence degree is greater than the preset feature coincidence degree, an identification image feature is obtained. The identification image feature is subjected to multi-scale transformation. Multi-scale transformation is a commonly used technique in image processing and computer vision, which aims to represent or transform images at multiple scales in order to more comprehensively capture information and features in images. Multi-scale transformation can help process targets of different scales in images, while improving the detection capability of features of different sizes. Common multi-scale transformation methods include image pyramid, scale space transformation, etc., and output high-dimensional mapping features.

[0069] The high-dimensional mapping features are used as corresponding features of the identification image features for image fusion processing, and the image features of the first image to be processed are output. The above method solves the technical problems of high feature extraction difficulty and low feature extraction quality when the image feature difference is small. By using multi-segmentation fusion technology for image feature extraction, the efficiency and quality of image feature extraction can be improved.

[0070] In one embodiment, as Figure 4The image feature extraction system provided by the application comprises a target identification module 1, a gray processing module 2, a pixel set acquisition module 3, a gray conversion module 4, an image feature extraction module 5, an image feature output module 6, wherein:

[0071] The target identification module 1 is used for identifying targets in the first image to be processed after the first image to be processed is acquired, and n targets are acquired, wherein n is a positive integer greater than or equal to 2.

[0072] The gray processing module 2 is used for performing gray processing on the first image to be processed and outputting a first gray image.

[0073] The pixel set acquisition module 3 is used for acquiring n sets of gray pixel sets of the n targets in the first gray image and n sets of remaining pixel sets in the first gray image, wherein the n sets of gray pixel sets correspond to the n sets of remaining pixel sets.

[0074] The gray conversion module 4 is used for inputting a set of gray pixel sets and a set of remaining pixel sets into a gray conversion model respectively for gray conversion, and outputting n sets of converted gray images.

[0075] The image feature extraction module 5 is used for outputting n segmentation images according to the n sets of gray images, and performing image feature extraction on the n segmentation images respectively, and outputting n image features.

[0076] The image feature output module 6 is used for outputting the image feature of the first image to be processed by performing image fusion processing on the n image features.

[0077] In one embodiment, the system further comprises:

[0078] A type identification module is used for identifying the types of the n targets in the first image to be processed and acquiring multiple types of targets.

[0079] A gray regulation channel setting module is used for setting gray regulation channels in the gray conversion model by the multiple types of targets, so that the number of the gray regulation channels is the same as the number of target types, wherein the gray regulation coefficients in each channel are different, and when the targets of the same type are input into the gray conversion model, the corresponding channels are matched for gray regulation.

[0080] In one embodiment, the system further comprises:

[0081] a gray data acquisition module, configured to acquire n groups of gray data and n groups of residual gray data according to the n groups of gray pixel sets and the n groups of residual pixel sets;

[0082] a gray image output module, configured to take the correspondence between the n groups of gray data and the n groups of residual gray data as a training data set, train based on the training data set with a target function, and output n groups of converted gray images when a response condition of the target function is met;

[0083] a target function module, wherein the target function is a function established to increase the difference between the n groups of gray data and the n groups of residual gray data.

[0084] In an embodiment, the system further comprises:

[0085] a gray image acquisition module, configured to acquire n groups of converted gray images, wherein the n groups of gray images are images obtained by performing gray conversion on the first to-be-processed image;

[0086] a pre-segmentation image output module, configured to perform edge recognition on the n groups of gray images through an edge algorithm, and output n pre-segmentation images;

[0087] a segmentation image output module, configured to perform edge smoothing processing on the n pre-segmentation images, and output n segmentation images.

[0088] In an embodiment, the system further comprises:

[0089] a target correlation generation module, configured to establish a connection relationship for each target corresponding to the n image features, and generate a target correlation;

[0090] a weight configuration module, configured to perform information entropy calculation on the target correlation, and configure a weight for feature fusion processing according to the information entropy calculation result;

[0091] an image fusion processing module, configured to perform image fusion processing on the n image features according to the configured weight, and output an image feature of the first to-be-processed image.

[0092] In an embodiment, the system further comprises:

[0093] An image feature acquisition module is configured to determine a feature coincidence degree between the n image features, and acquire an identification image feature when the feature coincidence degree is greater than a preset feature coincidence degree.

[0094] A high-dimensional mapping feature output module is configured to perform multi-scale transformation on the identification image feature and output a high-dimensional mapping feature.

[0095] An image fusion processing module is configured to perform image fusion processing on the high-dimensional mapping feature as a corresponding feature of the identification image feature.

[0096] In summary, the multi-division fusion image feature extraction method and system provided by the present application has the following technical effects:

[0097] 1. The technical problem of high difficulty and low quality of feature extraction when the difference between image features is small is solved, and the efficiency and quality of image feature extraction are improved by using the multi-division fusion technology for image feature extraction.

[0098] 2. By obtaining n sets of residual pixel sets, the features are highlighted by changing the gray value of the background, so that the image feature extraction becomes easier, thereby improving the accuracy and efficiency of image feature extraction. By constructing a gray conversion model, adaptive gray conversion can be performed according to the target type, thereby improving the quality of image gray conversion and reducing the difficulty of image feature extraction.

[0099] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0100] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for extracting image features through multi-segment fusion, characterized in that, The method includes: After acquiring the first image to be processed, the targets in the first image to be processed are identified to obtain n targets, where n is a positive integer greater than or equal to 2; The first image to be processed is processed to grayscale, and the first grayscale image is output. Obtain n sets of grayscale pixels for the n targets in the first grayscale image, and n sets of remaining pixels in the first grayscale image, wherein the n sets of grayscale pixels correspond to the n sets of remaining pixels; A set of grayscale pixels and a set of remaining pixels are respectively input into the grayscale conversion model for grayscale conversion, and the output is n sets of grayscale images after conversion; Based on the n sets of grayscale images, output n segmented images, and extract image features from each of the n segmented images to output n image features; By performing image fusion processing on the n image features, the image features of the first image to be processed are output; After obtaining n targets, the method further includes: Type recognition is performed on n targets in the first image to be processed to obtain multiple target categories; The grayscale adjustment channels in the grayscale conversion model are set according to the multiple target types, so that the number of grayscale adjustment channels is the same as the number of target types. The grayscale adjustment coefficient in each channel is different. When the same type of target is input into the grayscale conversion model, the corresponding channel is matched for grayscale adjustment. The method further includes: Based on the n sets of grayscale pixels and the n sets of remaining pixels, obtain n sets of grayscale data and n sets of remaining grayscale data; The correspondence between the n sets of grayscale data and the n sets of remaining grayscale data is used as the training data set. The training is performed based on the training data set with an objective function. When the response condition of the objective function is satisfied, the n sets of converted grayscale images are output. The objective function is a function established with the goal of increasing the difference between the n sets of grayscale data and the n sets of remaining grayscale data.

2. The method as described in claim 1, characterized in that, Based on the n sets of grayscale images, output n segmented images, including: Obtain n sets of grayscale images after conversion, wherein the n sets of grayscale images are images obtained by converting the first image to be processed into grayscale; Edge detection is performed on the n groups of grayscale images using an edge algorithm, and n pre-segmented images are output. The n pre-segmented images are processed to smooth the edges, and the n segmented images are output.

3. The method as described in claim 1, characterized in that, The method further includes: performing image fusion processing on the n image features to output the image features of the first image to be processed; Establish connection relationships between the targets corresponding to the n image features to generate target correlation; Information entropy is calculated based on the target correlation, and the information entropy calculation result is configured as a weight during feature fusion processing; Based on the configured weights, the n image features are then subjected to image fusion processing to output the image features of the first image to be processed.

4. The method as described in claim 3, characterized in that, The method further includes: Determine the feature overlap degree among the n image features; when the feature overlap degree is greater than a preset feature overlap degree, obtain the identifier image features. The features of the identified image are subjected to multi-scale transformation to output high-dimensional mapping features; The high-dimensional mapping features are used as the corresponding features of the identified image features for image fusion processing.

5. A multi-segment fusion image feature extraction system, characterized in that, The system comprises the following steps for performing any one of the image feature extraction methods according to claims 1-4: The target recognition module is used to identify targets in the first image to be processed after acquiring the first image to be processed, and to acquire n targets, where n is a positive integer greater than or equal to 2; A grayscale processing module is used to perform grayscale processing on the first image to be processed and output a first grayscale image. A pixel set acquisition module is used to acquire n sets of grayscale pixel sets of the n targets in the first grayscale image, and n sets of remaining pixel sets in the first grayscale image, wherein the n sets of grayscale pixel sets correspond to the n sets of remaining pixel sets. A grayscale conversion module is used to input a set of grayscale pixels and a set of remaining pixels into a grayscale conversion model for grayscale conversion, and output n sets of grayscale images after conversion. An image feature extraction module is used to output n segmented images based on the n sets of grayscale images, and to extract image features from the n segmented images respectively, outputting n image features; An image feature output module is used to output the image features of the first image to be processed by performing image fusion processing on the n image features.

Images