GB_ISAR Image Quality Assessment Method Based on Gray-Level Co-occurrence Matrix Spectrum

Abstract

This invention discloses a GB / T 1SAR image quality assessment method based on the gray-level co-occurrence matrix spectrum, belonging to the field of image processing technology. It solves the problems of existing technologies, such as difficulty in accurately estimating noise levels and inability to comprehensively assess the quality of GB / T 1SAR images. The method includes: acquiring and processing GB / T 1SAR images; converting the processed images into a grayscale image set, and processing each image to obtain a discrete image set; then performing multi-domain feature extraction to obtain the spatial and frequency domain features of each grayscale image; fusing the spatial and frequency domain features to obtain the image features corresponding to each grayscale image; and classifying the image features using a trained classification model to obtain the classification result. This method achieves the fusion of texture features extracted from the spatial domain and features extracted from the frequency domain of GB / T 1SAR images, requiring less time, being simple to operate, and achieving high evaluation accuracy.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a GB_ISAR image quality assessment method and apparatus based on gray-level co-occurrence matrix spectrum. Background Technology

[0002] Ground-based inverse synthetic aperture radar (GB_ISAR) imaging technology plays an increasingly important role in many areas such as space station identification, anti-satellite operations, and strategic early warning. As target imaging technology matures and improves, the need for image quality assessment technology is becoming increasingly urgent.

[0003] GB / T image quality assessment primarily involves a comprehensive evaluation of image resolution, target discernibility, focus, and continuity. However, this comprehensive evaluation remains a challenging and unresolved issue. Both GB / T imaging algorithm research and GB / T image quality assessment require an objective and impartial evaluation system. GB / T image quality assessment is a type of referenceless image quality assessment. Existing image quality assessment criteria, such as peak signal-to-noise ratio (PSNR) and root mean square error (RMSE), require a reference image. Furthermore, existing GB / T image quality assessment algorithms are mostly weighted comprehensive evaluation algorithms based on image entropy, contrast, and energy, which lack universality for GB / T images. Therefore, GB / T image quality assessment has always been a difficult problem, and research on radar image quality assessment is extremely limited.

[0004] The Gray-Level Co-occurrence Matrix (GLCM) is a statistical tool used to describe the texture features of an image. It is based on the relationships between pixels, statistically analyzing the frequency of occurrence of pixel pairs at different gray levels in an image. GLCM was first proposed by Robert M. Haraldlick et al. in 1973, and subsequent research defined a series of statistical features to describe texture, such as Angular Second Moment, Contrast, Correlation, and Entropy. These features can be used to quantify the texture characteristics of different images.

[0005] Natural scene images possess content-independent statistical properties, such as the MSCN coefficients, which exhibit a Gaussian distribution. These properties are compromised when images are distorted and can be used to measure the degree of distortion. In existing techniques, Mittal et al. proposed the BRISQUE no-reference image quality assessment algorithm, which extracts quality-sensitive features, fits the MSCN coefficient distribution to a generalized Gaussian distribution model, and learns the mapping relationship between features and subjective ratings through a classification model, thereby assessing the perceived image quality. Another algorithm proposed by Moorthy in 2011 utilizes wavelet domain pyramid decomposition of images to calculate the Natural Scene Statistical (NSS) properties, and maps quality-sensitive features to the perceived image quality rating through a classification model. Other methods include using CNNs to extract texture features and inputting them into a support vector machine classification model for ISAR image quality assessment.

[0006] Existing ISAR image quality assessment methods mainly include target detection, signal-to-noise ratio (SNR), resolution, and natural image processing algorithms. However, target edge detection algorithms are sensitive to noise and may produce erroneous or false edges in complex situations. SNR is limited by statistical methods, making it difficult to accurately estimate the noise level of complex ISAR images. Resolution is often affected by hardware and system parameters, making it difficult to comprehensively reflect image quality. Natural image processing algorithms are sensitive to specific models or characteristics and may not be applicable to certain ISAR image distortions. Multi-index weighted synthesis methods rely on subjective judgment and reliable indicators, which may lead to unstable evaluation results. The gray-level co-occurrence matrix method is mainly applicable to optical images, while phase information in GB-ISAR images is indispensable for quality assessment. In summary, these methods are helpful in GB-ISAR image quality assessment but also have limitations. Summary of the Invention

[0007] This invention provides a GB_ISAR image quality assessment method based on gray-level co-occurrence matrix spectrum, which solves the problems of difficulty in accurately estimating noise levels and the inability to completely assess the quality of GB_ISAR images in the prior art. It achieves the fusion of features extracted from the spatial domain and features extracted from the frequency domain of GB_ISAR images, and is simple to operate with high assessment accuracy.

[0008] In a first aspect, the present invention provides a GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum, the method comprising: A GB_ISAR image is acquired, and the GB_ISAR image is preprocessed and then initially classified to obtain a processed image. The processed image is converted into a grayscale image set, and the grayscale level of each image in the grayscale image set is quantized into a set of discrete grayscale values ​​to obtain a discrete image set. Spatial domain features are extracted from each grayscale image in the discrete image set to obtain the spatial domain features of each grayscale image, and frequency domain features are extracted from each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image. The spatial domain features and the frequency domain features are fused to obtain the image features corresponding to each grayscale image; The image features are classified using a trained classification model to obtain the classification result.

[0009] In conjunction with the first aspect, in one possible implementation, the step of preprocessing the GB_ISAR image before performing initial image classification to obtain a processed image includes: The GB_ISAR images are sequentially labeled and classified to obtain the first training set; The first training set is cropped, scaled, mirrored, rotated, and its contrast and brightness are adjusted in combination to determine the second training set. The second training set is used for initial image classification to obtain the processed images.

[0010] In conjunction with the first aspect, in one possible implementation, the step of extracting spatial features from each grayscale image in the discrete image set to obtain the spatial features of each grayscale image includes: Create an initial gray-level co-occurrence matrix; Find the number of pixel pairs that meet the positional relationship and generation direction in each grayscale image in the discrete image set, and accumulate the number of pixel pairs at the corresponding element in the initial grayscale co-occurrence matrix to update the initial grayscale co-occurrence matrix to obtain the updated grayscale co-occurrence matrix; The spatial features are calculated based on the updated gray-level co-occurrence matrix to obtain the spatial features of each gray-level image; wherein, the spatial features include: the spatial energy, spatial contrast, spatial entropy and spatial correlation of the gray-level image corresponding to the updated gray-level co-occurrence matrix.

[0011] In conjunction with the first aspect, in one possible implementation, the step of extracting frequency domain features from each grayscale image in the discrete image set to obtain frequency domain features includes: By displacing pixels in each grayscale image in the discrete image set by different distances, multiple pixel pairs corresponding to each pixel are obtained; The frequency of occurrence of each pixel pair in the discrete image set is counted to obtain the frequency of occurrence of each pixel pair and the total number of pixel pairs; The occurrence count of each pixel pair is normalized to obtain a normalized value; The normalized values ​​are transformed to the frequency domain, and the frequency domain features of each grayscale image are calculated.

[0012] In conjunction with the first aspect, in one possible implementation, the frequency domain features include: frequency domain energy, frequency domain contrast, frequency domain entropy, frequency domain correlation, and frequency domain entropy.

[0013] In conjunction with the first aspect, one possible implementation also includes: training the classification model to obtain a trained classification model, specifically including: Build a classification model; The image features corresponding to each image in the training set are input into the classification model for training. The trained classification model is used to obtain the classification result corresponding to each image. The classification result is compared with the label of each image in the training set to obtain the accuracy curve; Determine whether the accuracy curve is the same as the set curve. If yes, output the trained classification model; otherwise, perform cross-validation until the accuracy curve is the same as the set curve, and then output the trained classification model.

[0014] In a second aspect, the present invention provides a GB_ISAR image quality assessment device based on gray-level co-occurrence matrix spectrum, the device comprising: The preprocessing module is used to acquire GB_ISAR images, perform preprocessing on the GB_ISAR images, and then perform initial image classification to obtain the processed images. The discrete image set acquisition module is used to convert the processed image into a grayscale image set, and quantize the grayscale level of each image in the grayscale image set into a set of discrete grayscale values ​​to obtain a discrete image set. The feature extraction module is used to extract spatial domain features from each grayscale image in the discrete image set to obtain the spatial domain features of each grayscale image, and to extract frequency domain features from each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image. The feature fusion module is used to fuse the spatial domain features and the frequency domain features to obtain the image features corresponding to each grayscale image; The output module is used to classify the image features using the trained classification model to obtain the classification result.

[0015] Thirdly, the present invention provides a GB_ISAR image quality assessment server for gray-level co-occurrence matrix spectrum, the server including a memory and a processor; The memory is used to store computer-executable instructions; The processor is used to execute the computer-executable instructions to implement the GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum.

[0016] Fourthly, the present invention provides a computer-readable storage medium having executable instructions, wherein a computer executing the executable instructions is capable of a GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum.

[0017] One or more technical solutions provided in this invention have at least the following technical effects or advantages: (1) The GB-ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum of multi-domain features proposed in this invention achieves accurate image classification by extracting multi-domain features of GB-ISAR images. The multi-domain features extract information from different spatial and frequency domains, which can capture richer and more diverse image features. (2) By extracting features from GB_ISAR images in multiple domains, this invention can better adapt to different environments and situations, and improve the robustness and generalization ability of the model. The effective fusion of features extracted from different domains can better express the overall information of the image and help to comprehensively consider the key features in different domains, especially the phase information of GB_ISAR images, thereby improving the quality of overall feature representation. Since images may exhibit diversity in different domains, feature extraction in a single domain may not cover all situations. Multi-domain feature extraction helps to cope with the diversity of image data and improve the adaptability of the model. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments of the present invention or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the steps of a GB_ISAR image quality assessment method based on gray-level co-occurrence matrix spectrum provided in this embodiment of the invention; Figure 2 This is the basic structure of the "one-to-one" algorithm provided in the embodiments of the present invention; Figure 3 This invention provides a detailed breakdown of the basic structure of the "one-to-one" algorithm in embodiments of the present invention. Figure 4 A flowchart of a specific embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] Inverse synthetic aperture radar (ISAR) imaging technology plays an increasingly important role in many aspects such as target identification, strategic missile defense, anti-satellite defense, and strategic early warning. As target imaging technology gradually matures and improves, the technical demand for GB_ISAR image quality assessment is becoming more and more urgent.

[0022] GB_ISAR image quality assessment mainly involves a comprehensive evaluation of image resolution, the discernibility of targets in the image, the degree of image focus, and the continuity of the image. This work presents a very challenging and unresolved problem. Whether in the research of ISAR imaging algorithms, the acceptance of ISAR imaging projects, or the application of established imaging systems, effective image quality evaluation methods are needed to participate in algorithm considerations and imaging effect evaluation. Relying solely on human judgment often fails to yield objective and fair conclusions. While some progress has been made in evaluating the quality of optical and SAR images, and established evaluation systems are already in use, these methods, while valuable for GB_ISAR image evaluation, cannot be directly applied to GB_ISAR image evaluation due to the many characteristics of GB_ISAR images that differ from synthetic aperture radar (SAR) and optical images. Targeted evaluation methods based on the specific characteristics of GB_ISAR images are required. GB_ISAR image quality assessment is a non-parameter-based image quality assessment. Therefore, considering the special characteristics of GB_ISAR images, this invention proposes a batch processing quality evaluation method for GB_ISAR images, referencing SAR and optical image quality evaluation methods. Different datasets are established for different GB_ISAR images, eliminating the need for processing a large number of images. During dataset construction, image enhancement techniques such as cropping, rotation, scaling, and mirroring were combined with adjustments to brightness and contrast to improve image quality, increase visual information, modify image features, and enhance image diversity and adaptability. Multiple variants were generated from the original images to increase the sample size and thus the dataset size. Measurable features and several structural features were extracted from the grayscale images used for evaluation. After data processing and feature fusion, a set of multidimensional feature vectors was generated. Support Vector Machines (SVMs) were then selected for model training and classification to accurately assess GB_ISAR image quality. An accuracy curve was incorporated into the accuracy testing of the evaluation method. For the evaluation of a specific class of GB_ISAR images, the applicability is presented intuitively, allowing for timely adjustments to the classifier parameters.

[0023] In a first aspect, the present invention provides a GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum, the method comprising the following steps S101 to S105.

[0024] S101, acquire GB_ISAR image, preprocess the GB_ISAR image and then perform initial image classification to obtain the processed image.

[0025] Specifically, in step S101, a GB_ISAR image is acquired, and the GB_ISAR image is preprocessed and then initially classified to obtain a processed image, including the following steps S1011 to S1013.

[0026] S1011, perform label classification on GB_ISAR images to obtain the first training set.

[0027] S1012, perform cropping, scaling, mirroring, rotation, and adjustment of contrast and brightness combination operations on the first training set in sequence to determine the second training set.

[0028] S1013, perform initial image classification on the second training set to obtain the processed images.

[0029] For example, due to the long wavelength and limited bandwidth of its emitted waves, radar typically exhibits a sparse and isolated distribution of target scattering centers when used for target detection and 2D GB_ISAR imaging. This distribution of scattering centers reflects the target's physical structural characteristics, such as structure, size, shape, and material. However, GB_ISAR imaging is limited by measurement conditions and non-cooperative target motion compensation algorithms, so the actual measured images may be susceptible to distortion artifacts, blurring, and other phenomena.

[0030] Specifically, the main factors affecting GB_ISAR imaging quality include the following: Errors and defocusing effects: Due to the long wavelength of radar transmitted waves, radar imaging systems are susceptible to defocusing effects. This can lead to energy leakage in areas outside the target in GB_ISAR images, which can impair image clarity and thus affect image quality.

[0031] False scattering points: When the scattering structure of a target has special characteristics, false scattering points may be formed on the image. These false scattering points do not represent the actual target, but they may confuse the process of analyzing and identifying the target.

[0032] Noise and Interference: GB_ISAR images are often affected by noise and horizontal stripe interference, which can reduce image sharpness and the reliability of target analysis.

[0033] Therefore, image classification based on image quality mainly considers the above three aspects: whether the physical structure of the target is clear, whether there are false scattering points, whether the focus is good, and whether there is noise and horizontal stripe interference. Taking all these aspects into account, GB_ISAR images are classified into four types: excellent, good, medium, and poor, with the number of images in each type being [number missing]. The classification criteria are as follows: Excellent means that the target's physical structure is clear, that is, there are no false scattering points, almost no noise or lateral stripe interference, and the focusing is good.

[0034] "Good" means that the target's physical structure is distinguishable, that is, there are no false scattering points, but there is a small amount of noise and lateral stripe interference, and the focusing is average.

[0035] In the middle stage, the target's physical structure is blurred, but the local structure is distinguishable. There are a few false scattering points or a small amount of noise and lateral stripe interference, resulting in poor focusing.

[0036] Poor quality means that the overall physical structure of the target is indistinguishable, the local structure is indistinguishable, there are a large number of false scattering points or a large amount of noise and lateral stripe interference, and the focusing is very poor.

[0037] First, the GB_ISAR images were manually classified according to the classification criteria to obtain the first training set.

[0038] Then, based on the first training set, the GB_ISAR images are subjected to a combination of operations such as cropping, scaling, mirroring, rotating, and adjusting contrast and brightness to obtain the second training set. Next, the second training set is labeled: "1" for "excellent" samples; "2" for "good" samples; "3" for "medium" samples; and "3" for "poor" samples.

[0039] S102, convert the processed image into a grayscale image set, and quantize the grayscale level of each image in the grayscale image set into a set of discrete grayscale values ​​to obtain a discrete image set.

[0040] Specifically, in step S102, firstly, the processed image (the size of the processed image is...) is... The gray levels are converted into a set of discrete values. If the processed image is not a grayscale image, it needs to be converted to a grayscale image, and the gray levels of the image need to be quantized. It is usually limited to between 0 and 255.

[0041] S103, perform spatial domain feature extraction on each grayscale image in the discrete image set to obtain the spatial domain features of each grayscale image, and perform frequency domain feature extraction on each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image.

[0042] Specifically, in step S103, spatial features are extracted from each grayscale image in the discrete image set to obtain the spatial features of each grayscale image, including the following steps S1031 to S1033.

[0043] S1031, Create the initial gray-level co-occurrence matrix.

[0044] S1032, find the number of pixel pairs that meet the positional relationship and generation direction on each grayscale image, and accumulate the number of pixel pairs at the corresponding element in the initial grayscale co-occurrence matrix, update the initial grayscale co-occurrence matrix, and obtain the updated grayscale co-occurrence matrix.

[0045] S1033, calculate the spatial features based on the updated gray-level co-occurrence matrix to obtain the spatial features of each gray-level image; wherein, the spatial features include: the spatial energy, spatial contrast, spatial entropy and spatial correlation of the gray-level image corresponding to the updated gray-level co-occurrence matrix.

[0046] Specifically, step S103 further includes: extracting frequency domain features from each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image, including the following steps.

[0047] (1) By displacing pixels in each grayscale image in the discrete image set by different distances, multiple pixel pairs are obtained for each pixel. Specifically, pixel pairs are created by moving pixels in the image, and each pixel pair consists of two adjacent pixels.

[0048] (2) Count the number of occurrences of each pixel pair in the discrete image set to obtain the number of occurrences of each pixel pair and the total number of pixel pairs. Specifically, the number of occurrences reflects the frequency of different pixel combinations in each image.

[0049] (3) Normalize the occurrence frequency of each pixel pair to obtain a normalized value. Specifically, normalization is performed to convert the original data into a standardized form, which facilitates comparison of the occurrence frequency of the same pixel pair in different images.

[0050] (4) Transform the normalized values ​​to the frequency domain and calculate the frequency domain features of each grayscale image. Specifically, the frequency domain features include: frequency domain energy, frequency domain contrast, frequency domain entropy, frequency domain correlation, and frequency domain entropy.

[0051] Here, a specific implementation example for extracting frequency domain features is provided.

[0052] The gray-level co-occurrence matrix spectrum represents the relative positions of different frequency components of an image in the frequency domain, which reflects the phase and amplitude information of the GB_ISAR image.

[0053] Step 1, assuming pixels Moving the image over different distances within a discrete image set will result in various combinations of pixel pairs. , Represents pixels Pixel values ​​when no movement has occurred. Represents pixels The pixel value at the moved position; specific combinations of multiple pixel pairs. Represented as: (1.1) in, This indicates the grayscale level.

[0054] Step two, Represents pixels In displacement After that, the pixel value and pixel value The probability of them occurring simultaneously Represented as: (1.2) in, Represents a pixel; Represents pixels Pixel values ​​when no movement has occurred; Represented as pixels The pixel value at the new position after the movement; Represents pixels The x-coordinate value of the position; express The ordinate value of the position; This represents the Kronecker delta function, which is 1 when both conditions are met, and 0 otherwise.

[0055] Step 3: For the discrete image set, count the frequency of each pixel pair combination. Let... Let be the number of times each pixel pair combination occurs in the discrete image set, then the total number of occurrences is: (1.3) in, This represents the number of gray levels in a discrete image set. This indicates the number of times each pixel pair combination occurs in the discrete image set.

[0056] Step 4: Count the number of occurrences of each pixel pair combination in the discrete image set. Normalization is performed by... This is achieved by dividing each element in the matrix by the total number of pixel pair combinations. Each element in Represented as: (1.4) In this invention, normalization ensures that comparisons between different images are comparable.

[0057] Step 5: Perform a two-dimensional discrete Fourier transform on the normalized elements obtained in Step 4. Specifically, perform a two-dimensional discrete Fourier transform on equation (1.4) to the frequency domain, as expressed by the formula: (1.5) in, Indicates the range frequency. Indicates the azimuth frequency. This represents the number of pixels at a distance from a discrete image set. This represents the number of pixels in the orientation direction of a discrete image set.

[0058] In this invention, step five, where phase information in the frequency domain reflects the relative positions of different frequency components in the image, involves first shifting each pixel in the discrete image set to obtain pixel pairs corresponding to each pixel; then, the pixel value is determined. and pixel value The probability of simultaneous occurrence; then, using the occurrence count of each pixel pair combination in the discrete image set, obtain the total occurrence count of each pixel pair combination in the discrete image set; then, calculate the occurrence count of each pixel pair combination in the discrete image set... Normalization is performed; then, the normalized elements are subjected to a two-dimensional discrete Fourier transform to obtain the frequency domain value corresponding to each pixel.

[0059] After normalizing the elements of the discrete image set and transforming them to the frequency domain, the frequency domain features in the processed image are extracted.

[0060] Calculate various statistical characteristics, such as amplitude spectrum, phase spectrum, energy, amplitude entropy, phase entropy, amplitude contrast, phase contrast, amplitude and phase correlation.

[0061] Amplitude: Amplitude represents the intensity of different frequency components in the frequency domain, expressed as... express: (1.6) in, Indicates the real part, Indicates the imaginary part; This represents the frequency domain value corresponding to a pixel.

[0062] Phase: Phase represents the phase information of different frequency components in a frequency range, and is expressed as... express: (1.7) Energy: Energy represents the intensity of different frequency components in the frequency domain, expressed as... express: (1.8) in, Indicates the frequency value in the range direction. Indicates the frequency value in the azimuth direction. This represents the frequency of a discrete image set in the range direction; This represents the frequency of a discrete image set in the azimuth direction. Indicates amplitude.

[0063] Contrast ratio: Calculates the contrast between two adjacent pixels. Contrast ratio is divided into phase contrast and amplitude contrast. Used to highlight the tonal variations in an image, phase contrast Used to analyze the texture or periodic structure of an image.

[0064] (1.9) (1.10) in, , , This represents the amplitude of the frequency domain value of the current pixel. This represents the amplitude of the frequency domain value of the pixels adjacent to the current pixel. The phase sum of the frequency domain values ​​of the current pixel This represents the phase of the frequency domain values ​​adjacent to the current pixel.

[0065] Amplitude contrast It is usually related to the difference in signal strength or amplitude, while phase contrast... It is related to the phase difference of the signal.

[0066] In some applications, amplitude contrast It can more intuitively reflect changes in the brightness or intensity of a signal, while phase contrast... Used to measure the phase information of a signal, such as the periodicity of a waveform.

[0067] Entropy: Entropy reflects the degree of information disorder in the frequency domain, and here it is divided into amplitude entropy. and phase entropy That is, the complexity of amplitude is expressed by amplitude entropy. The complexity of the phase distribution is represented by phase entropy. Indicates amplitude entropy and phase entropy The specific calculations are as follows: (1.11) (1.12) in, and All are constants.

[0068] Correlation: Correlation represents the correlation between different frequency components in the frequency domain, and can be expressed by calculating the correlation between the amplitude spectrum and the phase spectrum. (1.13) in, (1.14) in, This represents the mean of the amplitude. The mean of the phase. The standard deviation of amplitude, The standard deviation of the phase, The first part of the amplitude spectrum represents the... One element, The first phase spectrum One element, This indicates the number of samples. The correlation between the amplitude spectrum and the phase spectrum represents the texture features of the image; changes in the correlation can reflect the edge features and texture structure of the GB_ISAR image.

[0069] S104, fuse spatial domain features and frequency domain features to obtain the image features corresponding to each grayscale image.

[0070] S105: Use the trained classification model to classify the image features and obtain the classification result.

[0071] Specifically, the method provided by the present invention also includes training the classification model, specifically including the following steps S1051 to S1504.

[0072] S1051, Construct a classification model.

[0073] S1052, input the image features corresponding to each image in the training set into the classification model for training, and obtain the classification result corresponding to each image.

[0074] S1053, compare the classification result with the label of each image in the training set to obtain the accuracy curve.

[0075] S1054: Determine if the accuracy curve is the same as the set curve. If not, perform cross-validation until the accuracy curve is the same as the set curve, and output the trained classification model.

[0076] For example, classification models have many advantages in classification by defining a generative classifier by finding the maximum margin in the feature space. In step one of this invention, the image is classified into four types, which constitutes a multi-class problem. Using one-vs-one (OvO) encoding, the multi-class problem is transformed into multiple binary classification sub-problems, and multiple binary classification models are trained. These sub-problems are independent of each other, and each sub-problem's binary classification model is used to distinguish one class pair. During prediction, the outputs of all sub-classifiers are combined to determine the final multi-class classification result. This step is shown below: First, the texture features are input into the classification model. Specifically, for Each category, build This involves using a binary classification model, where these classifiers work together to classify multi-class problems. In this example, with four classes, six binary classifiers are built, each classifier distinguishing one class. For each class pair, a small training set is created containing samples belonging to both classes and labeled as positive (+1) and negative (-1). A binary classification model is then trained for each class pair.

[0077] Then, the classification model is trained using texture features to obtain the classification result. Specifically, each binary classification model is trained using its own training set. During prediction, for OvO encoding, the final prediction result of multiple binary classification models is determined by voting or other rules. A voting method is used, classifying the test sample using all trained binary classification models. The classifier adds one vote to the class it classifies, and the class with the most votes is ultimately identified as the output class of the test sample. "One-to-one" multi-classification algorithms rarely misclassify; even if a binary classifier misclassifies a sample, it usually does not affect the final judgment result. Figure 2 The diagram shows the basic four-class classification structure of the "one-to-one" algorithm. Figure 3 The basic structure of the "one-to-one" algorithm is further subdivided.

[0078] A second vote is taken based on the classification results to obtain the final voting results.

[0079] Finally, the type with the most votes is selected for output.

[0080] The method provided by this invention involves classifying GB_ISAR images using the human eye, and then merging GB_ISAR images for which no readily available dataset exists to construct a GB_ISAR image dataset. Next, a gray-level co-occurrence matrix is ​​used to extract texture features from the GB_ISAR images, including entropy, contrast, correlation, and energy. These extracted features are input into a support vector machine (classification model) classifier for image classification. The multi-class problem is broken down into multiple binary classification sub-problems, using a one-to-one encoding method, and then multiple binary classification models are trained. During prediction, voting or other rules can be used to determine the final prediction results of the multiple binary classification models.

[0081] This method has the following characteristics and advantages: Dataset Construction: The method allows for the creation of custom datasets by classifying and merging GB_ISAR images using human eyes, even when no existing dataset exists. This increases flexibility and enables the handling of problem-specific image classification tasks.

[0082] Image augmentation: Image augmentation of a dataset can improve the universality of images, thereby improving the performance of the model.

[0083] Rich Information Acquisition: The GB_ISAR image quality assessment method using the gray-level co-occurrence matrix spectrum proposed in this invention achieves accurate image classification by extracting multi-domain features from GB_ISAR images. These multi-domain features extract information from different spatial and frequency domains, enabling the capture of richer and more diverse image features. This improves the understanding and representation capabilities of GB_ISAR image content.

[0084] Adapting to different scenarios: By extracting features from GB_ISAR images in multiple domains, the algorithm can better adapt to different environments and situations, improving the robustness and generalization ability of the model.

[0085] Inter-domain information fusion: Effectively fusing features extracted from different domains can better express the overall information of an image, help to comprehensively consider key features in different domains, especially the phase information of GB_ISAR images, and improve the quality of overall feature representation.

[0086] Addressing diversity: Because images can exhibit diversity across different domains, feature extraction from a single domain may not be able to cover all situations. Multi-domain feature extraction helps address the diversity of image data and improves the adaptability of the model.

[0087] SVM multi-classifier: This method trains multiple binary SVM classifiers using a "one-to-one" approach. It features simple operation, high computational efficiency, and the ability to handle imbalanced data.

[0088] like Figure 4 As shown, Figure 4 This is a flowchart illustrating the steps of a specific embodiment of the present invention.

[0089] (1) From N GB_ISAR images, four categories of images are selected by human eye: "Excellent", "Good", "Medium", and "Poor". The number of images in each category is: ; (2) The four types of images classified in step (1) are cropped, scaled, rotated, mirrored, and their contrast and brightness are adjusted in combination to enhance the data, generate new variants from the original images, increase the diversity of images, and improve the universality of the dataset.

[0090] (3) The new images generated in step (2) are then classified, and the human eye selects four categories: "Excellent", "Good", "Average", and "Poor". The number of images in each category is [number missing]. ; and merge it with the images of the same type from step (1). After merging, the number of images in each type is . ; (4) The classified GB_ISAR images are labeled with different labels, namely “1” for “excellent”, “2” for “good”, “3” for “medium”, and “4” for “poor”, and the training set is constructed.

[0091] (5) From the remaining images, select four categories: "Excellent," "Good," "Average," and "Poor." The number of images in each category is [number missing]. ; (6) The GB_ISAR images selected in step (5) are labeled with different tags, namely, “1” for “excellent”, “2” for “good”, “3” for “medium”, and “4” for “poor”, and a test set is constructed. (7) Convert the images in the training set into grayscale images, use the grayscale images as input, and perform multi-domain feature extraction based on the grayscale co-occurrence matrix spectrum of multi-domain features. The extracted image features are u1.

[0092] (8) Input the extracted image features u1 into the support vector machine classification model for training to obtain the trained classification model.

[0093] (9) Compare the actual classification results of the training samples with the labels of the training samples and draw the accuracy curve.

[0094] (10) If the accuracy curve does not reach the ideal value, that is, the accuracy curve is different from the set curve, then cross-validation is performed to obtain the final trained classification model.

[0095] (11) Convert the images in the test set into grayscale images, take the grayscale images as input, and perform multi-domain feature extraction based on the grayscale co-occurrence matrix spectrum of multi-domain features. The extracted image features are u2.

[0096] (12) The image features u2 extracted in step (11) are put into the final trained classification model for classification to obtain the batch imaging quality evaluation results of GB_ISAR.

[0097] Secondly, the present invention provides a GB_ISAR image quality assessment device based on gray-level co-occurrence matrix spectrum, the device comprising: a preprocessing module, a discrete image set acquisition module, a feature extraction module, a feature fusion module, and an output module.

[0098] The preprocessing module is used to acquire GB_ISAR images, perform preprocessing on the GB_ISAR images, and then perform initial image classification to obtain the processed images.

[0099] The discrete image set acquisition module is used to convert the processed image into a grayscale image set, and quantize the grayscale level of each image in the grayscale image set into a set of discrete grayscale values ​​to obtain the discrete image set.

[0100] The feature extraction module is used to extract spatial features from each grayscale image in the discrete image set to obtain the spatial features of each grayscale image, and to extract frequency features from each grayscale image in the discrete image set to obtain the frequency features of each grayscale image.

[0101] The feature fusion module is used to fuse spatial domain features and frequency domain features to obtain the image features corresponding to each grayscale image.

[0102] The output module is used to classify image features using the trained classification model and obtain the classification results.

[0103] The apparatus or module described in the above embodiments can be implemented by a computer chip or physical entity, or by a product with a certain function. For ease of description, the above apparatus is described by dividing it into various modules according to their functions. In implementing this invention, the functions of each module can be implemented in one or more software and / or hardware. Of course, a module that implements a certain function can also be implemented by combining multiple sub-modules or sub-units.

[0104] The methods, apparatus, or modules described in this invention can be implemented in a computer-readable program code manner. The controller can be implemented in any suitable manner, for example, as a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of a memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code manner, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included within it for implementing various functions can also be considered as structures within the hardware component. Alternatively, the device used to implement various functions can be viewed as either a software module that implements the method or a structure within a hardware component.

[0105] Some modules in the apparatus described in this invention can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0106] This invention provides a GB_ISAR image quality assessment server based on gray-level co-occurrence matrix spectrum. The server includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes the computer-executable instructions to implement the GB_ISAR image quality assessment method based on gray-level co-occurrence matrix spectrum.

[0107] This invention provides a computer-readable storage medium having executable instructions, and a GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum when the computer executes the executable instructions.

[0108] The aforementioned storage media include, but are not limited to, random access memory (RAM), read-only memory (ROM), cache, hard disk drive (HDD), or memory card. The memory can be used to store computer program instructions.

[0109] While this invention provides the method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in this embodiment is merely one possible execution order among many and does not represent the only possible execution order. In actual device or client product execution, the methods shown in this embodiment or the accompanying drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0110] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, or it can be embodied in the process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0111] The various embodiments described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. All or part of this invention can be used in numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0112] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the present invention.

Claims

1. A GB_ISAR image quality assessment method based on gray-level co-occurrence matrix spectrum, characterized in that, include: A GB_ISAR image is acquired, and the GB_ISAR image is preprocessed and then initially classified to obtain a processed image. The processed image is converted into a grayscale image set, and the grayscale level of each image in the grayscale image set is quantized into a set of discrete grayscale values ​​to obtain a discrete image set. Spatial domain features are extracted from each grayscale image in the discrete image set to obtain the spatial domain features of each grayscale image. Frequency domain features are also extracted from each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image. The step of extracting spatial domain features from each grayscale image in the discrete image set to obtain the spatial domain features of each grayscale image includes: creating an initial grayscale co-occurrence matrix; finding the number of pixel pairs in each grayscale image in the discrete image set that satisfy the positional relationship and generation direction, and accumulating the number of pixel pairs at the corresponding element in the initial grayscale co-occurrence matrix; updating the initial grayscale co-occurrence matrix to obtain an updated grayscale co-occurrence matrix; and calculating the spatial domain features based on the updated grayscale co-occurrence matrix to obtain the spatial domain features of each grayscale image. The spatial domain features include: the spatial energy, spatial contrast, spatial entropy, and spatial correlation of the grayscale image corresponding to the updated grayscale co-occurrence matrix. The step of extracting frequency domain features from each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image includes: shifting pixels in each grayscale image in the discrete image set by different distances to obtain multiple pixel pairs corresponding to each pixel; counting the occurrences of each pixel pair in the discrete image set to obtain the occurrence count of each pixel pair and the total number of pixel pairs; normalizing the occurrence count of each pixel pair to obtain a normalized value; and transforming the normalized value to the frequency domain to calculate the frequency domain features of each grayscale image. The spatial domain features and the frequency domain features are fused to obtain the image features corresponding to each grayscale image; The image features are classified using a trained classification model to obtain the classification result.

2. The GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum according to claim 1, characterized in that, The process of preprocessing the GB_ISAR image followed by initial image classification to obtain the processed image includes: The GB_ISAR images are labeled and classified to obtain the first training set; The first training set is subjected to cropping, scaling, mirroring, rotation, and a combination of contrast and brightness adjustment operations in sequence to determine the second training set. The second training set is used for initial image classification to obtain the processed images.

3. The GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum according to claim 1, characterized in that, The frequency domain features include: frequency domain energy, frequency domain contrast, frequency domain entropy, frequency domain correlation, and frequency domain entropy.

4. The GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum according to claim 1, characterized in that, Also includes: The classification model is trained to obtain a trained classification model, which specifically includes: Build a classification model; The image features corresponding to each image in the training set are input into the classification model for training to obtain the classification result corresponding to each image; The classification result is compared with the label of each image in the training set to obtain the accuracy curve; Determine whether the accuracy curve is the same as the set curve. If yes, output the trained classification model; otherwise, perform cross-validation on the images in the training set until the accuracy curve is the same as the set curve, and output the trained classification model.

5. A GB_ISAR image quality assessment device for gray-level co-occurrence matrix spectrum, used to implement the method according to any one of claims 1 to 4, characterized in that, include: The preprocessing module is used to acquire GB_ISAR images, perform preprocessing on the GB_ISAR images, and then perform initial image classification to obtain the processed images. The discrete image set acquisition module is used to convert the processed image into a grayscale image set, and quantize the grayscale level of each image in the grayscale image set into a set of discrete grayscale values ​​to obtain a discrete image set. The feature extraction module is used to extract spatial domain features from each grayscale image in the discrete image set to obtain the spatial domain features of each grayscale image, and to extract frequency domain features from each grayscale image in the discrete image set to obtain the frequency domain features of each grayscale image. The feature fusion module is used to fuse the spatial domain features and the frequency domain features to obtain the image features corresponding to each grayscale image; The output module is used to classify the image features using the trained classification model to obtain the classification result.

6. A GB_ISAR image quality assessment server for gray-level co-occurrence matrix spectrum, characterized in that, Including memory and processor; The memory is used to store computer-executable instructions; The processor is used to execute the computer-executable instructions to implement the GB_ISAR image quality assessment method based on the gray-level co-occurrence matrix spectrum as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium has executable instructions, and when the computer executes the executable instructions, it can implement the GB_ISAR image quality assessment method of gray-level co-occurrence matrix spectrum as described in any one of claims 1-4.

Images