Fast DCT frequency component selection method for image recognition tasks

Abstract

The invention relates to the field of radio orientation technology, and specifically discloses a fast DCT frequency component selection method for image recognition tasks. The method comprises: frequency component selection set statistics, image recognition accuracy index determination, selection factor matching and image recognition feedback, obtaining an image to be processed, importing a frequency conversion component to obtain frequency components, completing fast DCT selection and statistics as a frequency component selection set, using the set to perform image task recognition, determining its recognition accuracy index, comparing it with a predefined recognition accuracy reference index to obtain a comparison result, matching selection factors based on the result so that the frequency components can adapt to different task conditions, combining image recognition effectiveness data, comprehensively evaluating the image recognition effectiveness value, and providing feedback on image recognition of the frequency component selection set. The whole process aims to optimize the efficiency and accuracy of the frequency component selection set in the image recognition task.

Description

Technical Field

[0001] The invention relates to the field of radio orientation technology, in particular to a fast DCT frequency component selection method for image recognition tasks. Background Art

[0002] DCT is a mathematical transformation widely used in the field of image and video compression. It converts time domain signals (such as image pixel values) into frequency domain signals (such as frequency components), which can effectively remove redundant information and achieve data compression. At the same time, the basic principle of the DCT algorithm is to decompose the input signal into a weighted sum of a series of orthogonal cosine basis functions. These basis functions have different frequencies, so the different frequency components in the signal can be separated. In the image recognition task, not all frequency components have an important influence on the recognition results. Therefore, selecting appropriate frequency components for subsequent processing is the key to improving recognition efficiency and accuracy.

[0003] For example, the invention patent with announcement number CN108710115B announces a method for detecting ground penetrating radar targets based on wavelet decomposition, which includes the following steps: the ground penetrating radar uses two frequencies to detect the same target and obtain two sets of echo data; the two sets of echo data are processed by MATLAB respectively to form a high-frequency B-scan image and a low-frequency B-scan image; the high-frequency B-scan image and the low-frequency B-scan image are decomposed by wavelets respectively to extract the horizontal high-frequency component of the high-frequency B-scan image and the vertical high-frequency component of the low-frequency B-scan image; the horizontal high-frequency component in step three is energy accumulated in the survey line direction and then energy threshold screening is performed to obtain the horizontal position information image of the target, and the vertical high-frequency component in step three is energy accumulated in the time direction and then energy threshold screening is performed to obtain the depth position information image of the target; step five, the horizontal position and depth position information images of the target are fused to obtain the position image of the target.

[0004] For example, the invention patent with announcement number CN111708028B is a SAR image secondary imaging method based on multifractal spectrum, which includes the following steps: performing two-dimensional local delayed autocorrelation analysis or pseudo-Wigner time-frequency expansion on SAR image pixels to obtain a four-dimensional array, the four-dimensional array including original independent variables and two-dimensional delayed components (or two-dimensional frequency components); retaining the original independent variables, performing two-dimensional multifractal spectrum analysis on the two-dimensional delay components (or two-dimensional frequency components) to obtain the multifractal spectrum of the local point of the SAR image; traversing each spatial point of the original SAR image, processing the multifractal spectrum, and obtaining the multifractal spectrum SAR secondary imaging result of the SAR image.

[0005] Combining the above technical solutions, it is found that there is a frequency component selection technical solution, and its frequency component selection method is relatively single, which leads to limited actual application scenarios of frequency component selection, and then affects the technical problem of classification accuracy, and ultimately affects the completion of image recognition tasks. Summary of the invention

[0006] In view of the deficiencies of the prior art, the present invention provides a fast DCT frequency component selection method for image recognition tasks, which can effectively solve the problems involved in the above-mentioned background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a fast DCT frequency component selection method for image recognition tasks, including: frequency component selection set statistics: obtaining an image to be processed and importing it into a frequency transformation component, thereby performing fast DCT frequency component selection on each frequency component, and finally counting each frequency component that has completed fast DCT frequency component selection, and recording it as a frequency component selection set; image recognition accuracy index determination: performing task recognition on the image according to the frequency component selection set, and determining the image recognition accuracy index of the frequency component selection set; selection factor matching: comparing the image recognition accuracy index of the frequency component selection set with a predefined image recognition accuracy reference index to obtain a comparison result, and matching the selection factor of the frequency component selection set based on the comparison result; image recognition feedback: obtaining image recognition effectiveness data of the frequency component selection set, combining the selection factors of the frequency component selection set, and comprehensively evaluating the image recognition effectiveness value of the frequency component selection set, thereby providing feedback on the image recognition of the frequency component selection set.

[0008] As a further method, the frequency component selection set, the specific analysis process is:

[0009] The frequency conversion component specifically includes a color space conversion subcomponent, a discrete cosine transform subcomponent, a frequency recombination subcomponent and a connection subcomponent; the color space conversion subcomponent specifically converts the image to be processed into a grayscale image, and the image to be processed is a color image; the discrete cosine transform subcomponent specifically divides the grayscale image to obtain grayscale sub-images, performs DCT (discrete cosine transform) on each grayscale sub-image, and obtains the coefficient matrix of each grayscale sub-image; the frequency recombination subcomponent recombines the coefficient matrix of each grayscale sub-image to obtain the frequency component, specifically combines the coefficients located at the same position in the coefficient matrix of each grayscale sub-image into a separate frequency component to complete the fast DCT frequency component selection; the connection subcomponent specifically counts the frequency components that have completed the fast DCT frequency component selection, connects the frequency components according to the frequency dimension, and obtains the frequency component selection set.

[0010] As a further method, the image recognition accuracy index of the frequency component selection set is determined, and the specific analysis process is: obtaining the image recognition data of the frequency component selection set, specifically including the signal power mean of each frequency component of the frequency component selection set within the recognition period, the noise power mean of each frequency component of the frequency component selection set within the recognition period, the actual mean of the fundamental frequency of each frequency component of the frequency component selection set within the recognition period, and the signal energy mean of each frequency component of the frequency component selection set within the recognition period; obtaining the total signal energy value of the image; performing ratio processing on the signal power mean of each frequency component of the frequency component selection set within the recognition period and the noise power mean of each frequency component of the frequency component selection set within the recognition period to obtain the signal-to-noise ratio mean of each frequency component of the frequency component selection set within the recognition period; performing ratio processing on the signal energy mean of each frequency component of the frequency component selection set within the recognition period and the total signal energy value of the image to obtain the signal-to-noise ratio of each frequency component of the frequency component selection set within the recognition period. The proportion of the mean signal energy within the component operation cycle; obtain the number of data signal losses of the color space transformation subcomponent within the component operation cycle, the number of data signal losses of the discrete cosine transformation subcomponent within the component operation cycle, the number of data signal losses of the frequency recombination subcomponent within the component operation cycle, the maximum electromagnetic intensity value of the color space transformation subcomponent within the operation cycle, the maximum electromagnetic intensity value of the discrete cosine transformation subcomponent within the operation cycle, and the maximum electromagnetic intensity value of the frequency recombination subcomponent within the operation cycle, and comprehensively analyze to obtain the information loss degree coefficient of the subcomponent; extract the reference value of the fundamental frequency from the image recognition information library; according to the mean signal-to-noise ratio of each frequency component of the frequency component selection set within the recognition cycle, the proportion of the mean signal energy of each frequency component of the frequency component selection set within the recognition cycle, the actual mean of the fundamental frequency of each frequency component of the frequency component selection set within the recognition cycle, the reference value of the fundamental frequency and the information loss degree coefficient of the subcomponent, comprehensively analyze to obtain the image recognition accuracy index of the frequency component selection set.

[0011] As a further method, the selection factors of the frequency component selection set are matched based on the comparison results. The specific matching process is: if the comparison result shows the first comparison result, the image recognition accuracy index of the frequency component selection set is matched with the selection factors corresponding to each predefined image recognition accuracy index interval, so as to match the selection factors of the frequency component selection set and record them as the information entropy of the frequency component selection set.

[0012] As a further method, the selection factor of the frequency component selection set is matched based on the comparison result, and also includes: if the comparison result shows a second comparison result, the image recognition accuracy index of the frequency component selection set is differenced with the image recognition accuracy reference index to obtain the image recognition deviation value of the frequency component selection set, and the image recognition deviation value is matched with the selection factor corresponding to each predefined image recognition deviation value interval, so as to obtain the selection factor of the frequency component selection set and record it as the recognition adjustment coefficient of the frequency component selection set.

[0013] As a further method, the image recognition effectiveness value of the frequency component selection set is comprehensively evaluated. The specific evaluation process is as follows: the sine wave frequency value of each frequency component in the frequency component selection set is subjected to standard deviation processing to obtain the variation degree coefficient of the frequency component in the frequency component selection set; the number of frequency components of non-zero elements in the frequency component selection set is subjected to ratio processing to the total number of frequency components in the frequency component selection set to obtain the sparsity degree coefficient of the frequency components in the frequency component selection set; according to the variation degree coefficient of the frequency components in the frequency component selection set, the sparsity degree coefficient of the frequency components in the frequency component selection set and the selection factor of the frequency component selection set, the image recognition effectiveness value of the frequency component selection set is obtained. The specific method is as follows:

[0014] ;

[0015] in ;

[0016] In the formula, The image recognition effectiveness value for the frequency component selection set, For the frequency component, select the coefficient of variation of the concentrated frequency component. Select the sparsity coefficient of the concentrated frequency component for the frequency component, is the selection factor for the frequency component selection set, The image recognition accuracy index for the selected set of frequency components, To provide accurate reference indicators for image recognition, is the information entropy of the selected set of frequency components, Identification adjustment coefficients for the selected set of frequency components, is the weight factor corresponding to the predefined degree of change coefficient in the image recognition information library, is the weight factor corresponding to the predefined sparsity coefficient in the image recognition information library, is the weight factor corresponding to the predefined selection factor in the image recognition information library, and e is a natural constant.

[0017] As a further method, the image recognition of the frequency component selection set is fed back, and the specific feedback process is: the image recognition effectiveness value of the frequency component selection set is verified with a predefined image recognition effectiveness preset value to generate an image recognition evaluation label. If the image recognition effectiveness value of the frequency component selection set is greater than the image recognition effectiveness preset value, the image recognition evaluation label is defined as an image recognition effectiveness label, and there is no need to provide feedback on the image recognition of the frequency component selection set; if the image recognition effectiveness value of the frequency component selection set is less than or equal to the image recognition effectiveness preset value, the image recognition evaluation label is defined as an image recognition abnormality label, and an early warning is issued for the image recognition of the frequency component selection set.

[0018] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0019] (1) The present invention provides a fast DCT frequency component selection method for image recognition tasks, obtains an image to be processed, imports the frequency components obtained by the frequency transform component, conveniently selects and processes the frequency components, completes the fast DCT selection and counts them as a frequency component selection set, uses the set to perform image task recognition, determines its recognition accuracy index, and compares the comparison result with the predefined recognition accuracy reference index. Based on the result matching selection factor, the frequency component can adapt to different task conditions, thereby improving the flexibility and adaptability of the frequency component selection set. Combined with the image recognition effectiveness data, the image recognition effectiveness value is comprehensively evaluated, and the image recognition of the frequency component selection set is fed back. The whole process is aimed at optimizing the efficiency and accuracy of the frequency component selection set in the image recognition task;

[0020] (2) The present invention performs task recognition on an image based on a frequency component selection set, wherein the frequency component selection set is customized according to specific features of the image, and the most important frequency components for image recognition are selected. This targeted selection helps to reduce the interference of noise and redundant information, thereby improving the accuracy of the frequency component selection set in image recognition. By determining the image recognition accuracy index of the frequency component selection set, the performance of the frequency component can be more intuitively understood;

[0021] (3) The present invention obtains the image recognition effectiveness data of the frequency component selection set and combines it with the selection factors of the frequency component selection set to comprehensively evaluate the image recognition effectiveness value of the frequency component selection set, which is used to measure the actual effect of the frequency component selection set in the image recognition task, and helps to more accurately understand whether the currently selected frequency component meets the recognition requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention is further described with the aid of the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other accompanying drawings based on the following accompanying drawings without creative work.

[0023] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0024] Figure 2 Schematic diagram of discrete cosine transform;

[0025] Figure 3 This is a schematic diagram of frequency reorganization;

[0026] Figure 4 This is a schematic diagram of the frequency component connection. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] Reference Figure 1 As shown, the present invention provides a fast DCT frequency component selection method for image recognition tasks, including: frequency component selection set statistics: obtaining the image to be processed and importing it into the frequency transformation component, thereby performing fast DCT frequency component selection on each frequency component, and finally counting the frequency components that have completed the fast DCT frequency component selection and recording them as the frequency component selection set.

[0029] It should be explained that fast DCT frequency component selection, where DCT is a mathematical transformation that converts time domain signals, such as image pixel values ​​or video frames, into frequency domain signals, namely frequency components. Through DCT, redundant information in the signal can be effectively removed to achieve data compression. In image processing, DCT is usually used to decompose an image into components of different frequencies, where low-frequency components represent the main structure and smooth areas of the image, while high-frequency components represent the details and edges of the image. Frequency components refer to the components of a signal at different frequencies.

[0030] In image processing, not all frequency components are necessary. By selectively retaining or removing certain frequency components, image compression, denoising, enhancement and other processing can be achieved. The purpose of fast DCT frequency component selection is to reduce the number of frequency components that need to be processed as much as possible while ensuring image quality, thereby improving processing speed and efficiency.

[0031] Specifically, the frequency component selection set has a specific analysis process as follows:

[0032] The frequency conversion component specifically includes a color space conversion subcomponent, a discrete cosine transformation subcomponent, a frequency recombination subcomponent and a connection subcomponent.

[0033] The color space conversion subcomponent specifically converts the image to be processed into a grayscale image, and the image to be processed is a color image.

[0034] The input of the frequency conversion component is image data represented based on the RGB color space. , where H represents the height of the image, W represents the width of the image, and C represents the number of channels of the image. Convert it to YCrCb color space and output , where X represents the color space and R represents the image data matrix.

[0035] The above YCrCb color space decomposes color information into two components: brightness (Y) and chrominance (Cr and Cb). Among them, Y: brightness component, which indicates the brightness of the pixel and is related to the light intensity of the image. It is equivalent to grayscale or brightness information and is the main component of the image; Cr: chromaticity red component, which indicates the color difference between red and brightness and is used to adjust the red component; Cb: chromaticity blue component, which indicates the color difference between blue and brightness and is used to adjust the blue component.

[0036] The discrete cosine transform subcomponent specifically divides the grayscale image to obtain grayscale sub-images, performs DCT (discrete cosine transform) on each grayscale sub-image, and obtains a coefficient matrix of each grayscale sub-image.

[0037] The above discrete cosine transform subcomponents, such as Figure 2 As shown, Figure 2 is a schematic diagram of discrete cosine transform, Figure 2 It can be seen that discrete cosine transform is specifically to transform It is divided into three channels: Y, Cr, and Cb, which are divided into ( ) 8×8 image blocks, perform discrete cosine transform (DCT) on each image block of each channel, and obtain ( ) 8×8 DCT coefficient matrices, where is the representation image of channel Y, as well as is the representation of the brightness component, as well as is the brightness component coefficient of the image block, Represented as a coefficient matrix of an image block.

[0038] In a specific embodiment, when the input spatial domain image size is When , the size of the output data is ( ).

[0039] The frequency recombining subcomponent recombines the frequency components from the coefficient matrices of the grayscale sub-images, specifically, the coefficients at the same position in the coefficient matrices of the grayscale sub-images are combined into separate frequency components to complete the fast DCT frequency component selection.

[0040] The above frequency recombinant subcomponents, such as Figure 3 As shown, Figure 3 is a schematic diagram of frequency reorganization, Figure 3 It can be seen that the DCT coefficient matrix obtained for each channel is reorganized by combining the coefficients at the same position in different DCT coefficient matrices into separate frequency components. Each channel in Y, Cr, and Cb obtains 64 independent frequency components, among which , , Represented as the DCT coefficient matrix corresponding to each image block, , , It represents the 0th, tth, and 63rd frequency components obtained by frequency reorganization of the DCT coefficient matrix of the Y channel. Concat represents frequency reorganization. represents the frequency component for which fast DCT frequency component selection is performed for channel Y, is the output of the frequency conversion component, that is, the image input The corresponding frequency domain data.

[0041] The connection subcomponent specifically counts the frequency components that have completed the fast DCT frequency component selection, connects the frequency components according to the frequency dimension, and obtains a frequency component selection set.

[0042] The above-mentioned connection subcomponents, such as Figure 4 As shown, Figure 4 is a schematic diagram of frequency component connection, Figure 4 It can be seen that the frequency components of Y, Cr, and Cb are connected according to the frequency dimension to obtain the frequency data ,in , That is the output of the frequency conversion component, the frequency component selection set, , , They respectively represent the frequency components for completing fast DCT frequency component selection for channel Y, channel Cr, and channel Cb.

[0043] In a specific embodiment, according to Figure 3 , Figure 4 The following process can be constructed:

[0044] Extract an image from the image recognition task, which is represented as , input it into the frequency conversion component to obtain its frequency representation ,Depend on Figure 3 , Figure 4 It can be seen that the construction formula is:

[0045] ;

[0046] ;

[0047] ;

[0048] in , ;

[0049] ;

[0050] ;

[0051] in, , Represents each 8×8 image block, Indicates the position of each image block The pixel value at That is, the DCT coefficient matrix corresponding to each image block, Indicates the position of each DCT coefficient matrix The DCT coefficient value at It represents the tth frequency component obtained after frequency reorganization of the DCT coefficient matrix of the Y channel. Indicates the frequency component corresponding to the Y channel, whose size is 64×( ), the processing of Cr and Cb channels is the same as Y, Represents image input The corresponding frequency domain data.

[0052] The frequency component selection set refers to those signal frequency components that play a key role in the image recognition task. In a specific embodiment, the image recognition task may be to recognize a face in an image, and the frequency component selection set may be signal frequency components of facial features such as frequency components of the eye area, frequency components of the nose area, and frequency components of the mouth area.

[0053] Image recognition accuracy index determination: Perform task recognition on the image according to the frequency component selection set, and determine the image recognition accuracy index of the frequency component selection set.

[0054] Furthermore, the image recognition accuracy index of the determination frequency component selection set is specifically analyzed as follows:

[0055] The image recognition data of the frequency component selection set is obtained, specifically including the signal power mean of each frequency component of the frequency component selection set within the recognition period, the noise power mean of each frequency component of the frequency component selection set within the recognition period, the actual mean of the fundamental frequency of each frequency component of the frequency component selection set within the recognition period, and the signal energy mean of each frequency component of the frequency component selection set within the recognition period.

[0056] Get the total signal energy value of the image.

[0057] The above-mentioned signal power can be obtained by squaring the coefficient of each frequency component in the coefficient matrix to calculate the signal power of each frequency component, and the noise power can be obtained by filter detection; the fundamental frequency can be measured by a spectrum analyzer; the signal energy value can be calculated by the amplitude and phase information corresponding to each frequency component. The energy is equal to the square of the spectrum amplitude. The spectrum amplitude can be measured by a spectrum analyzer. The total signal energy value of the image can be obtained by calculating the sum of the squares of all pixel grayscale values.

[0058] The above-mentioned recognition period specifically refers to the duration of time used for the frequency component selection set to perform the recognition task on the image. The recognition period is determined by the task execution manager based on a comprehensive analysis of factors such as the frequency component distribution status of the frequency component selection set, actual needs, and image quality.

[0059] The signal power mean of each frequency component of the frequency component selection set within the identification period is ratio-processed with the noise power mean of each frequency component of the frequency component selection set within the identification period to obtain the signal-to-noise ratio mean of each frequency component of the frequency component selection set within the identification period.

[0060] The signal energy mean of each frequency component of the frequency component selection set within the identification period is processed by ratio with the total signal energy value of the image to obtain the signal energy mean ratio of each frequency component of the frequency component selection set within the identification period.

[0061] The number of data signal losses of the color space transformation subcomponent during the component operation cycle, the number of data signal losses of the discrete cosine transformation subcomponent during the component operation cycle, the number of data signal losses of the frequency recombination subcomponent during the component operation cycle, the maximum electromagnetic intensity value of the color space transformation subcomponent during the operation cycle, the maximum electromagnetic intensity value of the discrete cosine transformation subcomponent during the operation cycle, and the maximum electromagnetic intensity value of the frequency recombination subcomponent during the operation cycle are obtained, and the information loss degree coefficient of the subcomponent is obtained through comprehensive analysis.

[0062] The above-mentioned data signal loss times can be extracted from the log record report of each sub-component, and the electromagnetic intensity value can be detected by a fluxmeter.

[0063] The above-mentioned component operation cycle is specifically a period of time used by each subcomponent to perform fast DCT frequency component selection. The component operation cycle is determined by the task execution manager based on a comprehensive analysis of factors such as the operating status of each component, actual needs, and image quality.

[0064] The information loss coefficient of the above subcomponents is analyzed as follows:

[0065] ;

[0066] In the formula, is the information loss degree coefficient of the subcomponent, which is used in this embodiment to measure the specific degree of the negative impact caused by the data signal loss and the electromagnetic intensity interference of each subcomponent itself in the process of performing the image recognition task of the frequency component selection set. The more the data signal is lost and the greater the electromagnetic intensity of the subcomponent itself, the greater the negative impact on the image recognition accuracy index of the frequency component selection set;

[0067] The number of data signal losses of the color space conversion subcomponent during the component operation cycle refers to the number of times or amounts of signal information loss or distortion caused by the failure of signal data to be correctly transmitted or stored due to equipment failure, data transmission error, algorithm defect, etc. during the operation cycle of the color space conversion subcomponent. This number directly reflects the performance stability and reliability of the color space conversion subcomponent. If the number of data signal losses is too large, the accuracy and consistency of the frequency component selection set for image recognition will be affected;

[0068] The number of data signal losses of the discrete cosine transform subcomponent during the component operation cycle refers to the number of times or quantities of signal information loss or distortion caused by the failure of signal data to be correctly transmitted or stored due to equipment failure, etc. during the operation cycle of the discrete cosine transform subcomponent;

[0069] The number of data signal losses of the frequency recombination subassembly during the operation cycle of the frequency recombination subassembly refers to the number of times or quantities of signal information loss or distortion caused by the failure of signal data to be correctly transmitted or stored due to equipment failure, etc. during the operation cycle of the frequency recombination subassembly;

[0070] The maximum electromagnetic intensity value of the color space transformation subcomponent during the operation cycle refers to the maximum value that the electromagnetic radiation intensity of the color space transformation subcomponent can reach during the operation process. This electromagnetic intensity will interfere with the signal data and affect the recognition of the image by the frequency component selection set;

[0071] is the maximum electromagnetic intensity value of the discrete cosine transform subcomponent during the operation cycle, which refers to the maximum value that the electromagnetic radiation intensity of the discrete cosine transform subcomponent itself can reach during the operation process;

[0072] The maximum electromagnetic intensity value of the frequency recombination subassembly during its operation cycle refers to the maximum value that the electromagnetic radiation intensity of the frequency recombination subassembly can reach during its operation;

[0073] It is the weight factor corresponding to the number of data signal loss predefined in the image recognition information library.

[0074] In a specific embodiment, the weight factor corresponding to the number of data signal loss times can be directly obtained from the image recognition information library when in use. The factor represents the numerical value of the degree of influence on the information loss degree coefficient of the sub-component during the information loss degree analysis of the sub-component. The corresponding relationship can be a pre-set mapping relationship. For example, the number of data signal loss times and the weight factors corresponding to the number of data signal loss times preset in the image recognition information library form a mapping set, and the real-time number of data signal loss times is brought into the mapping set to obtain the weight factor corresponding to the number of data signal loss times. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1).

[0075] is the weight factor corresponding to the maximum electromagnetic intensity value predefined in the image recognition information library, and e is a natural constant.

[0076] In a specific embodiment, the weight factor corresponding to the maximum electromagnetic intensity value can be directly obtained from the image recognition information library during use. The factor represents the numerical value of the influence on the information loss degree coefficient of the sub-component during the information loss degree analysis of the sub-component. The corresponding relationship can be a pre-set mapping relationship. For example, the maximum electromagnetic intensity value and the weight factor corresponding to the preset maximum electromagnetic intensity value in the image recognition information library form a mapping set, and the real-time maximum electromagnetic intensity value is brought into the mapping set to obtain the weight factor corresponding to the maximum electromagnetic intensity value. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1).

[0077] In this embodiment, if the color space transformation subcomponent, the discrete cosine transformation subcomponent and the frequency recombination subcomponent have higher electromagnetic intensity values, it means that the electromagnetic radiation intensity of each component itself is too large. When the electromagnetic intensity increases, it may interfere with the normal operation of the color space transformation subcomponent, resulting in the loss or distortion of the data signal, thereby increasing the number of data signal losses. When the data signal of the color space transformation subcomponent is lost, some frequency component information in the image will be lost or distorted, which will affect the accuracy and completeness of the frequency component selection set, thereby reducing the accuracy and reliability of image recognition.

[0078] The reference value of the fundamental frequency is extracted from the image recognition information library.

[0079] According to the mean signal-to-noise ratio of each frequency component of the frequency component selection set in the recognition period, the mean signal energy proportion of each frequency component of the frequency component selection set in the recognition period, the actual mean fundamental frequency of each frequency component of the frequency component selection set in the recognition period, the reference value of the fundamental frequency and the information loss degree coefficient of the subcomponent, the image recognition accuracy index of the frequency component selection set is obtained. The specific method is as follows:

[0080] ;

[0081] In the formula, is the image recognition accuracy index of the frequency component selection set, which is used in this embodiment to measure the accuracy and performance of the frequency component selection set when performing the image recognition task. The larger the image recognition accuracy index is, the more obvious the performance of the frequency component selection set in image recognition is;

[0082] is the information loss degree coefficient of the subcomponent, which represents the information loss degree coefficient of the subcomponent obtained by comprehensive analysis of the number of data signal losses of the color space transformation subcomponent during the component operation cycle, the number of data signal losses of the discrete cosine transformation subcomponent during the component operation cycle, the number of data signal losses of the frequency recombination subcomponent during the component operation cycle, the maximum electromagnetic intensity value of the color space transformation subcomponent during the operation cycle, the maximum electromagnetic intensity value of the discrete cosine transformation subcomponent during the operation cycle, and the maximum electromagnetic intensity value of the frequency recombination subcomponent during the operation cycle;

[0083] is the mean signal-to-noise ratio of the pth frequency component of the frequency component selection set in the recognition period, which refers to the ratio of the signal strength to the noise strength of the pth frequency component of the frequency component selection set in the recognition period. A high signal-to-noise ratio usually means higher recognition accuracy and less false recognition rate.

[0084] p is the number of each frequency component, , L is the total amount of frequency components;

[0085] The ratio of the signal energy mean of the pth frequency component in the frequency component selection set within the recognition period refers to the ratio of the information that the frequency component may contain to the overall information of the image. It is used to measure the contribution of different frequency components to the overall signal energy or importance in the image recognition task.

[0086] The actual mean value of the fundamental frequency of the pth frequency component of the frequency component selection set within the identification period refers to the average value of the actual frequency values ​​that can be detected by the fundamental frequency of the frequency component within the identification period, where the fundamental frequency is the lowest frequency component in the signal;

[0087] is the reference value of the fundamental frequency, which refers to the predefined standard value of the fundamental frequency;

[0088] is the fundamental frequency deviation value of the pth frequency component of the frequency component selection set within the identification period;

[0089] It is the impact factor corresponding to the unit value of the information loss coefficient predefined in the image recognition information library.

[0090] In a specific embodiment, the influence factor corresponding to the unit value of the information loss degree coefficient can be directly obtained from the image recognition information library when used, and the factor represents the value of the influence degree on the image recognition accuracy index of the frequency component selection set in the process of image recognition accuracy analysis of the frequency component selection set, and the corresponding relationship can be a preset mapping relationship. For example, the information loss degree coefficient and the influence factor corresponding to the unit value of the information loss degree coefficient preset in the image recognition information library form a mapping set, and the real-time information loss degree coefficient is brought into the mapping set to obtain the influence factor corresponding to the unit value of the information loss degree coefficient. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1);

[0091] is the weight factor corresponding to the mean value of the signal-to-noise ratio predefined in the image recognition information library;

[0092] In a specific embodiment, the weight factor corresponding to the mean signal-to-noise ratio can be directly obtained from the image recognition information library when used, and the factor represents the numerical value of the degree of influence on the image recognition accuracy index of the frequency component selection set in the process of image recognition accuracy analysis of the frequency component selection set, and the corresponding relationship can be a pre-set mapping relationship. For example, the mean signal-to-noise ratio and the weight factor corresponding to the mean signal-to-noise ratio preset in the image recognition information library form a mapping set, and the real-time mean signal-to-noise ratio is brought into the mapping set to obtain the weight factor corresponding to the mean signal-to-noise ratio. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, the value range is (0, 1);

[0093] is the weight factor corresponding to the mean proportion of signal energy predefined in the image recognition information library;

[0094] In a specific embodiment, when used, the weight factor corresponding to the signal energy mean ratio can be directly obtained from the image recognition information library, and the factor represents the numerical value of the degree of influence on the image recognition accuracy index of the frequency component selection set in the process of image recognition accuracy analysis of the frequency component selection set, and the corresponding relationship can be a pre-set mapping relationship. For example, the signal energy mean ratio and the weight factor corresponding to the signal energy mean ratio preset in the image recognition information library form a mapping set, and the real-time signal energy mean ratio is brought into the mapping set to obtain the weight factor corresponding to the signal energy mean ratio. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1);

[0095] is the weight factor corresponding to the fundamental frequency deviation value predefined in the image recognition information library, and e is a natural constant.

[0096] In a specific embodiment, the weight factor corresponding to the fundamental frequency deviation value can be directly obtained from the image recognition information library when in use. The factor represents the numerical value of the degree of influence on the image recognition accuracy index of the frequency component selection set during the image recognition accuracy analysis of the frequency component selection set. The corresponding relationship can be a pre-set mapping relationship. For example, the fundamental frequency deviation value and the weight factor corresponding to the fundamental frequency deviation value preset in the image recognition information library form a mapping set, and the real-time fundamental frequency deviation value is brought into the mapping set to obtain the weight factor corresponding to the fundamental frequency deviation value. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1).

[0097] In signal processing, the fundamental frequency of the frequency component usually has the highest signal-to-noise ratio, because the fundamental frequency component of the frequency component is the main energy source of the signal and is relatively stable. When the signal-to-noise ratio of the frequency component is high, the fundamental frequency component is easier to be accurately identified and analyzed, so that the fundamental frequency component performs better and is easier to approach the reference value, reducing the fundamental frequency deviation and increasing the image recognition accuracy index of the frequency component selection set. At the same time, since the fundamental frequency component of the frequency component is the main energy source of the signal, the frequency component also shows more required information, so its signal energy ratio is also relatively high, which helps to analyze and optimize the quality and characteristics of the signal and improve the image recognition accuracy index of the frequency component selection set.

[0098] In this embodiment, when the degree of information loss of the sub-component is high, the feature information in the image may be severely lost after frequency conversion. The loss of feature information will cause a large error in the selection of frequency components in the frequency component selection set. Some important frequency components are ignored or misjudged due to information loss, making it impossible to accurately capture the main features and detail information of the image, thereby reducing the accuracy of image recognition.

[0099] Selection factor matching: The image recognition accuracy index of the frequency component selection set is compared with the predefined image recognition accuracy reference index to obtain a comparison result, and based on the comparison result, the selection factor of the frequency component selection set is matched.

[0100] Specifically, the image recognition accuracy index of the frequency component selection set is compared with the image recognition accuracy reference index predefined in the image recognition information library to obtain a comparison result, specifically a first comparison result or a second comparison result.

[0101] The first comparison result is that the image recognition accuracy index of the frequency component selection set is greater than or equal to the image recognition accuracy reference index.

[0102] The second comparison result is that the image recognition accuracy index of the frequency component selection set is less than the image recognition accuracy reference index.

[0103] Furthermore, the selection factors of the frequency component selection set are matched based on the comparison result, and the specific matching process is:

[0104] If the comparison result shows the first comparison result, the image recognition accuracy index of the frequency component selection set is matched with the selection factors corresponding to each image recognition accuracy index interval predefined in the image recognition information library. The specific matching process is: obtain a mapping set of the image recognition accuracy index and the selection factor of the frequency component selection set from the image recognition information library, first determine the interval to which the image recognition accuracy index of the frequency component selection set belongs, and assign the selection factor corresponding to the interval to the frequency component selection set corresponding to the image recognition accuracy index, so as to match the selection factor of the frequency component selection set and record it as the information entropy of the frequency component selection set.

[0105] The information entropy of the above frequency component selection set is expressed as a value of the richness of the information contained in the frequency component selection set. The richer the information contained, the greater the contribution of the frequency component selection set to image recognition.

[0106] Specifically, the matching of the selection factors of the frequency component selection set based on the comparison result further includes:

[0107] If the comparison result shows the second comparison result, the image recognition accuracy index of the frequency component selection set is differenced with the image recognition accuracy reference index to obtain the image recognition deviation value of the frequency component selection set, and the image recognition deviation value is matched with the selection factors corresponding to each image recognition deviation value interval predefined in the image recognition information library. The specific matching process is: obtain a mapping set of image recognition deviation values ​​and selection factors of the frequency component selection set from the image recognition information library, first determine the interval to which the image recognition deviation value of the frequency component selection set belongs, and assign the selection factor corresponding to the interval to the frequency component selection set corresponding to the image recognition deviation value, so as to obtain the selection factor of the frequency component selection set by matching, and record it as the recognition adjustment coefficient of the frequency component selection set.

[0108] The recognition adjustment coefficient of the above-mentioned frequency component selection set refers to the coefficient used to adjust the weight of the frequency component selection set so that each frequency component of the frequency component selection set occupies a dominant position in the signal. Increasing the corresponding adjustment coefficient can make the frequency components in the frequency component selection set easier to identify and analyze, increase the detection accuracy of the frequency component selection set in the image recognition task, and reduce the negative impact of component noise.

[0109] In a specific embodiment, the matching of the selection factors can be designed into a selection factor layer, and its specific structure is as follows:

[0110] A selection factor layer, consisting of a length The vector of all 1s is called the selection factor. Each element in the selection factor corresponds to A frequency component in Multiplied by the selection factor, the selection factor is trained together with other required parameters, and the value of the selection factor is updated by back propagation. Each selection factor reflects the importance of its corresponding frequency component for the image recognition task.

[0111] For the training of the selection factor, an early stopping indicator is designed, called the selection factor vibration amplitude (SFVA), which will affect the training of the selection factor. When the value of SFVA is 0, the training is stopped, and the subscripts corresponding to the factors with the largest values ​​are selected from the trained selection factors. This subscript set is expressed as .

[0112] According to the obtained From frequency domain input Select the most critical K frequency components and express these K frequency components as , which is calculated as follows:

[0113] ;

[0114] in, Represents the index corresponding to the K maximum values ​​selected from the selection factor, Indicates based on from The most important frequency components are selected from Represents image input The corresponding frequency domain data.

[0115] Image recognition feedback: Obtain the image recognition effectiveness data of the frequency component selection set, combine it with the selection factors of the frequency component selection set, and comprehensively evaluate the image recognition effectiveness value of the frequency component selection set, so as to provide feedback on the image recognition of the frequency component selection set.

[0116] Furthermore, the image recognition validity data of the frequency component selection set specifically includes the sinusoidal wave frequency value of each frequency component in the frequency component selection set, the number of frequency components of non-zero elements in the frequency component selection set, and the total number of frequency components in the frequency component selection set.

[0117] The above-mentioned sine wave frequency value can be detected by a digital oscilloscope, and the number of frequency components of non-zero elements and the total number of frequency components can be extracted from the output report of the connected sub-component.

[0118] Specifically, the image recognition effectiveness value of the frequency component selection set is comprehensively evaluated, and the specific evaluation process is as follows:

[0119] The sine wave frequency value of each frequency component in the frequency component selection set is processed with standard deviation to obtain the coefficient of variation of the frequency components in the frequency component selection set.

[0120] The number of frequency components of non-zero elements in the frequency component selection set is ratioed to the total number of frequency components in the frequency component selection set to obtain the sparsity coefficient of the frequency components in the frequency component selection set.

[0121] According to the comprehensive evaluation of the change degree coefficient of the frequency components in the frequency component selection set, the sparse degree coefficient of the frequency components in the frequency component selection set and the selection factor of the frequency component selection set, the image recognition effectiveness value of the frequency component selection set is obtained. The specific method is as follows:

[0122] ;

[0123] in ;

[0124] In the formula, is the image recognition effectiveness value of the frequency component selection set, which is an indicator for measuring the contribution of the frequency component selection set in the image recognition task in this embodiment, indicating the performance of the frequency component selection set in the application. The larger the image recognition effectiveness value is, the more each frequency component in the frequency component selection set can reflect the real signal data of the image;

[0125] The coefficient of variation of the frequency components in the frequency component selection set refers to the discrete degree or variation range of the frequency component distribution in the frequency component selection set. If the sine wave frequency values ​​of the frequency components are very different, the standard deviation will be large, and the obtained frequency component variation coefficient will also be large, indicating that the frequency components in the selection set are more dispersed and have a larger variation range, so the frequency components will carry different information, which may be beneficial for image recognition tasks;

[0126] The sparsity coefficient of the frequency components in the frequency component selection set refers to how many proportions of frequency components in the frequency component selection set are non-zero. This coefficient reflects the sparsity of the frequency components in the frequency component selection set, that is, the proportion of effective frequency components in the total frequency components. In the frequency component selection set, the proportion of effective frequency components should be relatively high, while redundant or invalid frequency components should be effectively filtered out. Such sparse representation can improve the accuracy and efficiency of the image recognition process in the frequency component selection set;

[0127] is the image recognition accuracy index of the frequency component selection set, which means that the image recognition accuracy index of the frequency component selection set is obtained by comprehensive analysis of the mean signal-to-noise ratio of each frequency component of the frequency component selection set in the recognition period, the mean signal energy proportion of each frequency component of the frequency component selection set in the recognition period, the actual mean fundamental frequency of each frequency component of the frequency component selection set in the recognition period, the reference value of the fundamental frequency and the information loss degree coefficient of the subcomponent;

[0128] The image recognition accuracy reference index refers to the standard value corresponding to the predefined image recognition accuracy index;

[0129] is the selection factor for the frequency component selection set, is the information entropy of the selected set of frequency components, Identification adjustment coefficients for the selected set of frequency components;

[0130] The weight factor corresponding to the predefined change degree coefficient in the image recognition information library;

[0131] In a specific embodiment, the weight factor corresponding to the degree of change coefficient can be directly obtained from the image recognition information library when used, and the factor represents the numerical value of the degree of influence on the image recognition effectiveness value of the frequency component selection set during the image recognition effectiveness analysis of the frequency component selection set, and the corresponding relationship can be a preset mapping relationship, for example, the degree of change coefficient and the weight factor corresponding to the degree of change coefficient preset in the image recognition information library form a mapping set, and the real-time degree of change coefficient is brought into the mapping set to obtain the weight factor corresponding to the degree of change coefficient, wherein the mapping relationship can be a one-to-one correspondence or a many-to-one relationship, and its value range in this embodiment is (0, 1);

[0132] is a weight factor corresponding to a predefined sparsity coefficient in the image recognition information library;

[0133] In a specific embodiment, the weight factor corresponding to the sparsity coefficient can be directly obtained from the image recognition information library when used, and the factor represents the numerical value of the degree of influence on the image recognition effectiveness value of the frequency component selection set during the image recognition effectiveness analysis of the frequency component selection set, and the corresponding relationship can be a pre-set mapping relationship. For example, the sparsity coefficient and the weight factor corresponding to the preset sparsity coefficient in the image recognition information library form a mapping set, and the real-time sparsity coefficient is brought into the mapping set to obtain the weight factor corresponding to the sparsity coefficient. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1);

[0134] is the weight factor corresponding to the predefined selection factor in the image recognition information library, and e is a natural constant.

[0135] In a specific embodiment, the weight factor corresponding to the selection factor can be directly obtained from the image recognition information library when in use. The factor represents the numerical value of the degree of influence on the image recognition effectiveness value of the frequency component selection set during the image recognition effectiveness analysis of the frequency component selection set. The corresponding relationship can be a pre-set mapping relationship. For example, the selection factor and the weight factor corresponding to the preset selection factor in the image recognition information library form a mapping set, and the real-time selection factor is brought into the mapping set to obtain the weight factor corresponding to the selection factor. The mapping relationship can be a one-to-one correspondence or a many-to-one relationship. In this embodiment, its value range is (0, 1).

[0136] In this embodiment, when the degree of variation of the frequency components in the frequency component selection set is large, it may mean that the image or signal is more complex in details and requires more frequency components to represent it, which may lead to an increase in the sparsity coefficient because more non-zero elements are required to capture these details. On the contrary, when the degree of variation of the frequency components in the frequency component selection set is small, it may mean that the image or signal is simpler in details and can be represented by fewer frequency components, which may lead to a decrease in the sparsity coefficient because the proportion of effective frequency components is relatively low. Therefore, there is a mutual influence and connection between the degree of variation coefficient of the frequency components in the frequency component selection set and the sparsity coefficient, and they jointly describe the characteristics of the frequency component selection set.

[0137] In this embodiment, when the image recognition accuracy of the frequency component selection set is high and greater than or equal to the image recognition accuracy reference index, information entropy is used as a selection factor to emphasize the frequency components containing more information in the frequency component selection set, which helps to capture key details in the image, thereby improving the accuracy and robustness of recognition; when the recognition accuracy is low and less than the image recognition accuracy reference index, by introducing a recognition adjustment coefficient as a selection factor, the weight of the frequency component selection set can be adjusted to focus more on those frequency components that have a positive impact on the recognition results, which helps to reduce redundant information and improve recognition efficiency and speed. This method allows the selection factors and weights to be dynamically adjusted according to changes in recognition accuracy, thereby enhancing the adaptability of the frequency component selection set to different images and scenes, which helps to improve the generalization ability and robustness of the image recognition system in practical applications.

[0138] Furthermore, the image recognition of the frequency component selection set is fed back, and the specific feedback process is:

[0139] The image recognition effectiveness value of the frequency component selection set is verified with the predefined image recognition effectiveness preset value to generate an image recognition evaluation label. If the image recognition effectiveness value of the frequency component selection set is greater than the image recognition effectiveness preset value, the image recognition evaluation label is defined as the image recognition effectiveness label, and there is no need to provide feedback on the image recognition of the frequency component selection set.

[0140] If the image recognition effectiveness value of the frequency component selection set is less than or equal to the preset image recognition effectiveness value, the image recognition evaluation label is defined as an image recognition abnormality label, and an early warning is issued for the image recognition of the frequency component selection set.

[0141] Specifically, the early warning for image recognition of the frequency component selection set is specifically early warning for fast DCT frequency component selection of the frequency component selection set.

[0142] The purpose of the above-mentioned warning is to timely discover and deal with abnormalities or problems that may occur during the selection process, so as to ensure the accuracy and efficiency of the frequency component selection set in the image or signal processing process.

[0143] The above-mentioned abnormalities or problems may be, such as abnormal frequency component selection, component selection noise problem, frequency component sparsity and other problems. If there is an electromagnetic noise problem in the selection process, in order to reduce the impact of electromagnetic interference on the data signal, electromagnetic shielding and protection measures can be used to electromagnetically shield and protect the frequency conversion components to reduce the negative impact of electromagnetic interference in the fast DCT frequency component selection process.

[0144] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A fast DCT frequency component selection method for image recognition tasks, characterized in that: include: Frequency component selection set statistics: the image to be processed is obtained and imported into the frequency transform component, whereby each frequency component performs fast DCT frequency component selection, and finally the frequency components that have completed fast DCT frequency component selection are counted and recorded as the frequency component selection set; Image recognition accuracy index determination: perform task recognition on the image according to the frequency component selection set, and determine the image recognition accuracy index of the frequency component selection set; Selection factor matching: comparing the image recognition accuracy index of the frequency component selection set with the predefined image recognition accuracy reference index to obtain a comparison result, and matching the selection factor of the frequency component selection set based on the comparison result; Image recognition feedback: Obtain the image recognition effectiveness data of the frequency component selection set, combine it with the selection factors of the frequency component selection set, and comprehensively evaluate the image recognition effectiveness value of the frequency component selection set, so as to provide feedback on the image recognition of the frequency component selection set.

2. The fast DCT frequency component selection method for image recognition tasks according to claim 1, characterized in that: The specific analysis process of the frequency component selection set is as follows: The frequency conversion component specifically includes a color space conversion subcomponent, a discrete cosine transformation subcomponent, a frequency reorganization subcomponent and a connection subcomponent; The color space conversion subcomponent specifically converts the image to be processed into a grayscale image, and the image to be processed is a color image; The discrete cosine transform subcomponent specifically divides the grayscale image to obtain grayscale sub-images, performs DCT (discrete cosine transform) on each grayscale sub-image, and obtains a coefficient matrix of each grayscale sub-image; The frequency recombining subcomponent recombines the frequency components from the coefficient matrices of the grayscale sub-images, specifically, the coefficients at the same position in the coefficient matrices of the grayscale sub-images are combined into separate frequency components to complete the fast DCT frequency component selection; The connection subcomponent specifically counts the frequency components that have completed the fast DCT frequency component selection, connects the frequency components according to the frequency dimension, and obtains a frequency component selection set.

3. The fast DCT frequency component selection method for image recognition tasks according to claim 2, characterized in that: The image recognition accuracy index of the determination frequency component selection set is specifically analyzed as follows: Acquire image recognition data of the frequency component selection set, specifically including the signal power mean of each frequency component of the frequency component selection set within the recognition period, the noise power mean of each frequency component of the frequency component selection set within the recognition period, the actual mean of the fundamental frequency of each frequency component of the frequency component selection set within the recognition period, and the signal energy mean of each frequency component of the frequency component selection set within the recognition period; Get the total signal energy value of the image; Performing ratio processing on the mean value of signal power of each frequency component of the frequency component selection set within the identification period and the mean value of noise power of each frequency component of the frequency component selection set within the identification period, to obtain the mean value of signal-to-noise ratio of each frequency component of the frequency component selection set within the identification period; Performing ratio processing on the signal energy mean of each frequency component of the frequency component selection set within the identification period and the total signal energy value of the image to obtain the signal energy mean ratio of each frequency component of the frequency component selection set within the identification period; Obtain the number of data signal losses of the color space conversion subcomponent during the component operation cycle, the number of data signal losses of the discrete cosine conversion subcomponent during the component operation cycle, the number of data signal losses of the frequency recombination subcomponent during the component operation cycle, the maximum electromagnetic intensity value of the color space conversion subcomponent during the operation cycle, the maximum electromagnetic intensity value of the discrete cosine conversion subcomponent during the operation cycle, and the maximum electromagnetic intensity value of the frequency recombination subcomponent during the operation cycle, and comprehensively analyze to obtain the information loss degree coefficient of the subcomponent; Extracting a reference value of the fundamental frequency from an image recognition information database; Based on the comprehensive analysis of the mean signal-to-noise ratio of each frequency component of the frequency component selection set within the identification period, the mean signal energy proportion of each frequency component of the frequency component selection set within the identification period, the actual mean fundamental frequency of each frequency component of the frequency component selection set within the identification period, the reference value of the fundamental frequency and the information loss degree coefficient of the sub-component, the image recognition accuracy index of the frequency component selection set is obtained.

4. The fast DCT frequency component selection method for image recognition tasks according to claim 3, characterized in that: The image recognition accuracy index of the frequency component selection set is compared with a predefined image recognition accuracy reference index to obtain a comparison result, specifically a first comparison result or a second comparison result; The first comparison result is that the image recognition accuracy index of the frequency component selection set is greater than or equal to the image recognition accuracy reference index; The second comparison result is that the image recognition accuracy index of the frequency component selection set is less than the image recognition accuracy reference index.

5. The fast DCT frequency component selection method for image recognition tasks according to claim 1, characterized in that: Based on the comparison result, the selection factors of the frequency component selection set are matched, and the specific matching process is: If the comparison result shows the first comparison result, the image recognition accuracy index of the frequency component selection set is matched with the selection factors corresponding to each predefined image recognition accuracy index interval to match the selection factors of the frequency component selection set and record them as the information entropy of the frequency component selection set.

6. The fast DCT frequency component selection method for image recognition tasks according to claim 1, characterized in that: The matching of the selection factors of the frequency component selection set based on the comparison result also includes: If the comparison result shows the second comparison result, the image recognition accuracy index of the frequency component selection set is differenced with the image recognition accuracy reference index to obtain the image recognition deviation value of the frequency component selection set, and matched with the selection factors corresponding to each predefined image recognition deviation value interval to obtain the selection factor of the frequency component selection set, and recorded as the recognition adjustment coefficient of the frequency component selection set.

7. The fast DCT frequency component selection method for image recognition tasks according to claim 6, characterized in that: The image recognition validity data of the frequency component selection set specifically includes the sine wave frequency value of each frequency component in the frequency component selection set, the number of frequency components of non-zero elements in the frequency component selection set, and the total number of frequency components in the frequency component selection set.

8. The fast DCT frequency component selection method for image recognition tasks according to claim 7, characterized in that: The image recognition effectiveness value of the frequency component selection set is comprehensively evaluated, and the specific evaluation process is: Perform standard deviation processing on the sine wave frequency value of each frequency component in the frequency component selection set to obtain the coefficient of variation of the frequency component in the frequency component selection set; The number of frequency components of non-zero elements in the frequency component selection set is processed by ratio with the total number of frequency components in the frequency component selection set to obtain the sparsity coefficient of the frequency components in the frequency component selection set; According to the comprehensive evaluation of the change degree coefficient of the frequency components in the frequency component selection set, the sparse degree coefficient of the frequency components in the frequency component selection set and the selection factor of the frequency component selection set, the image recognition effectiveness value of the frequency component selection set is obtained. The specific method is as follows: ； in ; In the formula, The image recognition effectiveness value for the frequency component selection set, For the frequency component, select the coefficient of variation of the concentrated frequency component. Select the sparsity coefficient of the concentrated frequency component for the frequency component, is the selection factor for the frequency component selection set, The image recognition accuracy index for the selected set of frequency components, To provide accurate reference indicators for image recognition, is the information entropy of the selected set of frequency components, Identification adjustment coefficients for the selected set of frequency components, is the weight factor corresponding to the predefined degree of change coefficient in the image recognition information library, is the weight factor corresponding to the predefined sparsity coefficient in the image recognition information library, is the weight factor corresponding to the predefined selection factor in the image recognition information library, and e is a natural constant.

9. The fast DCT frequency component selection method for image recognition tasks according to claim 1, characterized in that: The feedback of the image recognition of the frequency component selection set is performed, and the specific feedback process is: The image recognition effectiveness value of the frequency component selection set is verified with a predefined image recognition effectiveness preset value to generate an image recognition evaluation label. If the image recognition effectiveness value of the frequency component selection set is greater than the image recognition effectiveness preset value, the image recognition evaluation label is defined as an image recognition effectiveness label, and there is no need to provide feedback on the image recognition of the frequency component selection set. If the image recognition effectiveness value of the frequency component selection set is less than or equal to the preset image recognition effectiveness value, the image recognition evaluation label is defined as an image recognition abnormality label, and an early warning is issued for the image recognition of the frequency component selection set.

10. The fast DCT frequency component selection method for image recognition tasks according to claim 9, characterized in that: The early warning for the image recognition of the frequency component selection set is specifically early warning for the fast DCT frequency component selection of the frequency component selection set.

Images