Image acquisition method and device based on multi-channel image acquisition card

Abstract

The invention relates to the technical field of image acquisition, and discloses an image acquisition method and device based on a multi-channel image acquisition card. The method comprises the following steps: inputting multiple paths of image signals into a multi-channel image acquisition card for signal conditioning to obtain digital image data; performing adaptive gain control on the digital image data to obtain gain-adjusted image data; carrying out digital filtering, image enhancement and feature extraction on the image data of which the gain is adjusted to obtain processed image data; performing definition evaluation, noise evaluation and dynamic range evaluation on the processed image data to obtain an image quality evaluation status word; and performing hierarchical storage on the processed image data according to the image quality evaluation status word to obtain distributed storage image data, thereby improving the real-time processing capability and data management efficiency of the multi-channel image acquisition card.

Description

Technical Field

[0001] The present invention relates to the field of image acquisition, and in particular to an image acquisition method and device based on a multi-channel image acquisition card. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, higher and higher requirements are placed on the real-time, reliability and acquisition quality of multi-channel image acquisition systems. Traditional image acquisition methods often use a single signal conditioning circuit and a fixed gain control method when processing multi-channel image signals, which leads to unstable image quality under different lighting conditions and is difficult to meet the acquisition requirements in complex industrial environments. In terms of image data storage management, existing technologies generally adopt a unified storage strategy and fail to perform graded processing based on image quality, resulting in a waste of storage resources and affecting the efficiency of subsequent image processing.

[0003] In the image feature extraction and processing stage, current technical solutions mostly use a single feature extraction method, which cannot take into account both the spatial information and semantic information of the image. In particular, important high-resolution feature information is easily lost during image downsampling. In addition, the existing feature extraction methods lack an effective feature enhancement mechanism, making it difficult to accurately capture multi-scale features and long-distance dependencies in the image, affecting the accuracy and reliability of image analysis. At the system implementation level, existing multi-channel image acquisition systems generally have problems such as insufficient real-time performance, a single quality assessment mechanism, and inefficient storage management. Especially when processing large-scale image data, due to the lack of a scientific quality assessment system and an intelligent storage management strategy, it is difficult to ensure the stable operation of the system and the effective management of data, which also increases the system's maintenance costs and resource consumption. Summary of the invention

[0004] The present invention provides an image acquisition method and device based on a multi-channel image acquisition card, and the present invention improves the real-time processing capability and data management efficiency of the multi-channel image acquisition card.

[0005] In a first aspect, the present invention provides an image acquisition method based on a multi-channel image acquisition card, the image acquisition method based on the multi-channel image acquisition card comprising: Inputting multi-channel image signals into a multi-channel image acquisition card for signal conditioning to obtain digital image data; Performing adaptive gain control on the digitized image data to obtain image data with adjusted gain; Performing digital filtering, image enhancement and feature extraction on the image data whose gain has been adjusted to obtain processed image data; Performing clarity assessment, noise assessment and dynamic range assessment on the processed image data to obtain an image quality assessment status word; The processed image data is hierarchically stored according to the image quality evaluation status word to obtain distributedly stored image data.

[0006] In a second aspect, the present invention provides an image acquisition device based on a multi-channel image acquisition card, the image acquisition device based on the multi-channel image acquisition card comprising: A signal conditioning module is used to input multiple image signals into a multi-channel image acquisition card for signal conditioning to obtain digital image data; A gain control module, used for performing adaptive gain control on the digitized image data to obtain image data with adjusted gain; A feature extraction module, used for performing digital filtering, image enhancement and feature extraction on the image data with adjusted gain to obtain processed image data; An evaluation module, used for performing clarity evaluation, noise evaluation and dynamic range evaluation on the processed image data to obtain an image quality evaluation status word; The hierarchical storage module is used to perform hierarchical storage on the processed image data according to the image quality evaluation status word to obtain distributed stored image data.

[0007] In the technical solution provided by the present invention, by designing an independent signal conditioning circuit and an adaptive gain control algorithm, high-precision acquisition and dynamic range adjustment of multi-channel image signals are achieved, and the quality and stability of image acquisition are effectively improved. Multi-level digital filtering and image enhancement processing are adopted, combined with feature extraction technology, to significantly improve the signal-to-noise ratio and clarity of the image, and a multi-dimensional image quality evaluation mechanism is innovatively introduced. Through the comprehensive evaluation of the three dimensions of clarity, noise and dynamic range, accurate quantification of image quality is achieved. A hierarchical storage strategy based on quality level is designed, and the storage efficiency of the system is optimized by rationally allocating storage resources, while ensuring the security and accessibility of important data. A dual-branch feature extraction network and a pooling feature enhancement module are adopted to effectively retain the spatial information and semantic information of the image, and improve the accuracy and robustness of feature extraction. By integrating an FPGA processing platform and a multi-level storage structure, efficient utilization of system resources is achieved, and the real-time processing capability and data management efficiency of the system are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0009] Figure 1A schematic diagram of a flow chart of an image acquisition method based on a multi-channel image acquisition card provided in an embodiment of the present application; Figure 2 A schematic block diagram of the structure of an image acquisition device based on a multi-channel image acquisition card provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0011] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0012] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0013] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0014] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0015] See also Figure 1 , Figure 1 A schematic diagram of a flow chart of an image acquisition method based on a multi-channel image acquisition card provided in an embodiment of the present application, such as Figure 1 As shown, the image acquisition method based on the multi-channel image acquisition card provided in the embodiment of the present application includes steps S100 to S500.

[0016] Step S100, inputting multiple image signals into a multi-channel image acquisition card for signal conditioning to obtain digital image data; It is understandable that the execution subject of the present invention may be an image acquisition device based on a multi-channel image acquisition card, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0017] Specifically, a programmable gain amplifier circuit is used to amplify the multi-channel image signals. The function of the programmable gain amplifier circuit is to provide sufficient gain to ensure that weak signals can be reliably collected, and to dynamically adjust the gain according to the amplitude characteristics of different input signals, so that the output signal is kept within a reasonable range, thereby preventing the occurrence of signal overload or insufficient amplitude. The amplified image signal is low-pass filtered by a second-order Butterworth filter circuit. The Butterworth filter has a flat frequency response characteristic, which can effectively suppress high-frequency noise without weakening the effective signal too much, and obtain a purer filtered image signal. The filtered image signal is input into a 14-bit high-speed AD conversion chip for analog-to-digital conversion. The converted digital signal reaches 2^14, that is, 16384 discrete levels, which can provide higher quantization accuracy. The digital signal is clock synchronized by a phase-locked loop (PLL) circuit to ensure that the digital signals of all channels are consistent in time, so that the subsequent multi-channel image data can be correctly aligned to obtain a multi-channel synchronized digital signal. The multi-channel synchronized digital signal is input into the data cache module inside the FPGA for cache processing. As a high-performance parallel processing device, FPGA (Field Programmable Gate Array) has significant advantages in real-time data processing and cache management. In order to improve the efficiency of data reading and storage, the data cache module adopts a dual-port RAM (Dual-Port RAM) structure, which allows different modules to read and write the same storage unit at the same time, improves data throughput and reduces storage delay, so that multi-channel data can still be efficiently accessed and managed under high-speed acquisition. The data format conversion of the cached digitized signal allows the original ADC output data to be reorganized to meet the predetermined image data format requirements, such as RGB, grayscale format or other customized formats, to obtain digital image data.

[0018] Step S200, performing adaptive gain control on the digitized image data to obtain image data with adjusted gain; Specifically, the digitized image data is subjected to histogram statistical analysis, the brightness distribution of each pixel is calculated, and the overall brightness value of each channel image is extracted. The number of pixels at different brightness levels is counted, and the average brightness is calculated to obtain a value representing the brightness level of the current image. The value is compared with the preset target brightness range, and the brightness difference between the two is calculated. The size of the brightness difference directly reflects the deviation between the current image and the expected brightness. The gain adjustment parameter is calculated by the proportional-integral control algorithm based on the brightness difference data to ensure that the gain adjustment can both quickly respond to brightness changes and maintain stability. Proportional-integral control (PI control) is an adaptive control method, in which the proportional term is used to quickly respond to brightness deviations, and the integral term is used to eliminate long-term deviations. The combined effect enables the system to smoothly converge to the target brightness value. According to the calculated gain control amount, the parameter for adjusting the image gain is obtained, but since the gain adjustment cannot change too quickly, in order to prevent sudden changes or flickering in the image, the gain control amount is slope-limited. The slope limit imposes constraints on the rate of gain adjustment so that the gain change does not exceed the set range within a certain period of time, thereby ensuring a smooth transition of the gain adjustment process and avoiding drastic changes that lead to image quality degradation. After slope limiting, the limiting gain parameters are obtained. The limited gain parameters are written into the parameter storage area inside the FPGA for subsequent gain search and application. The internal storage area of ​​the FPGA is used to efficiently store these gain parameters so that the system can access them at any time and dynamically adjust the image gain based on the latest calculated gain parameters. In order to improve the efficiency and real-time performance of gain correction, a gain lookup table is generated based on the stored gain parameter data. The lookup table quickly maps the input brightness value to the appropriate gain adjustment amount, avoids the computational overhead caused by real-time calculation, and improves the response speed of gain adjustment. After the gain lookup table is generated, the input digitized image data is gain corrected according to the table. By applying the corresponding gain adjustment amount to each pixel point, the final output image achieves the best matching effect in brightness, and the image data with adjusted gain is obtained.

[0019] Step S300, performing digital filtering, image enhancement and feature extraction on the image data whose gain has been adjusted to obtain processed image data; Specifically, the image data with adjusted gain is input into a fifth-order FIR filter for digital filtering. Due to its linear phase characteristics and stability, the FIR filter is applied to signal processing, especially in the field of image processing. Through a reasonably designed fifth-order FIR filter, high-frequency noise and interference signals in the image are effectively removed, while retaining useful edge information, so that the filtered image can reduce noise interference while maintaining clarity. The filtered image data is gamma corrected through a lookup table. The core of gamma correction is to use a lookup table (LUT) to perform nonlinear mapping of pixel values, so that the grayscale of the image is more in line with the visual perception characteristics of the human eye. By adjusting the gamma parameters, the local contrast of the image is enhanced, making the dark details more obvious, while avoiding overexposure of the highlight area, and obtaining the gamma-corrected image data. After completing the gamma correction, in order to extract the structural information in the image, the Sobel operator is used for edge detection. The Sobel operator is an edge detection method based on gradient operation. It identifies the edge position by calculating the horizontal and vertical gradient changes of the image grayscale. By performing the Sobel operator calculation on the gamma-corrected image data, the edge detection data is obtained. The data contains the main contour information in the image and can highlight the significant feature areas. Based on the edge detection data, the image is edge sharpened to enhance the detail information in the image, making the edge transition clearer and improving the overall visual quality. In the edge sharpening process, high-pass filtering or Laplace transform and other methods are used, combined with the output data of the Sobel operator, to enhance the gamma-corrected image, so that the edge information is more prominent, and the edge-enhanced image data is obtained. Feature detection is performed on the edge-enhanced image data to identify key points in the image, such as corners, texture features or significant areas. Using algorithms such as Harris corner detection, SIFT (Scale Invariant Feature Transform) or ORB (Oriented Fast Binary Descriptor), feature point data is extracted from the edge-enhanced image data. These feature points are helpful for subsequent image matching, target recognition, and deep learning analysis. The extracted feature point data and the edge-enhanced image data are stored or transmitted, and the feature point data and the edge-enhanced image data are efficiently transmitted through the data bus inside the FPGA, so that the subsequent image processing unit can directly access the processed image data.

[0020] Step S400, performing clarity assessment, noise assessment and dynamic range assessment on the processed image data to obtain an image quality assessment status word; Specifically, the energy gradient is calculated for the processed image data. Energy gradient is an important indicator for measuring image details and clarity. The sharpness of the image is quantified by calculating the average gradient value and gradient variance of the image. The gradient calculation is implemented through convolution operation. The algorithm includes Sobel operator, Prewitt operator or Laplacian operator, etc., which can effectively extract the edge information of the image. The average gradient value and gradient variance of the image are compared with the standard test image, and the clarity evaluation result of the image is obtained by comparing with the ideal image. An image with a large gradient variance means that the image has clear edges and rich details, while an image with a small gradient variance indicates that the image is blurred or the details are lost. Based on these comparison results, the clarity evaluation result is generated, and finally a 4-bit status code is formed to indicate the clarity level of the image. The status code reflects whether the image is clear enough and whether there is an obvious blurred area or detail loss. The signal-to-noise ratio of the processed image data is calculated by local mean filtering. The signal-to-noise ratio reflects the ratio between the useful signal and the noise. The local mean filter eliminates high-frequency noise by calculating the average value of the local area. After calculating the signal-to-noise ratio data, the data is compared with the preset noise threshold to determine the noise level of the image. If the signal-to-noise ratio is lower than the preset threshold, it means that the image contains high noise and needs further processing or repair. Through comparison, the noise evaluation result is obtained to form a 4-bit status code, which indicates the noise level of the image, such as clear image, noise interference, and high noise. The processed image data is subjected to histogram analysis, and the brightness distribution of the image is understood by counting the number of pixels at each gray level in the image. The histogram distribution data can reveal the coverage of useful information in the image. If the effective pixels of the image are distributed within the entire digital range (such as 0-255), it indicates that the dynamic range of the image is wide and can show more brightness levels and details. On the contrary, if the effective pixels are mainly concentrated in a certain range, it indicates that the dynamic range of the image is limited, and overexposure or underexposure occurs. On this basis, the effective pixel coverage in the histogram distribution is compared with the preset digital range to obtain the dynamic range evaluation result. Images with a wider dynamic range are usually of higher quality, can show more detailed information, and are suitable for high-precision analysis applications. According to the clarity evaluation results, noise evaluation results and dynamic range evaluation results, high-order and low-order status data are generated respectively. The high-order status data indicates the quality level of the image. This data is a 4-bit status code generated based on the clarity evaluation results, reflecting the clarity of the image. The low-order status data indicates the noise level and dynamic range of the image. This data is composed of a 4-bit status code generated by the noise evaluation results and the dynamic range evaluation results, reflecting the noise interference in the image and the width of the dynamic range. The high-order status data and the low-order status data are combined to finally generate the image quality evaluation status word.

[0021] Step S500: hierarchically store the processed image data according to the image quality evaluation status word to obtain distributedly stored image data.

[0022] Specifically, the image quality assessment status word is parsed to extract the image quality grade data. The upper 4 bits of the image quality assessment status word are used to determine the image quality grade, which reflects the comprehensive evaluation results of multiple aspects such as image clarity, noise level and dynamic range. The image quality assessment status word is parsed to assign a clear quality grade to each frame of the image, usually ranging from grade 1 to grade 4. A data index table is generated based on the processed image data, which contains the timestamp, quality grade and storage location information of the image. The function of the data index table is to provide an effective management mechanism for image data, so that the corresponding image file can be quickly located and retrieved in the subsequent storage and access process. The data index information of each frame of the image is recorded in the table, where the timestamp is used to mark the time of image acquisition, the quality grade identifies the quality level of the image, and the storage location information indicates the storage location of the image data. These information together constitute the image storage management system to ensure that the image data can be efficiently organized and classified during the storage process. Based on the image quality grade and the support of the data index table, the image data is stored in a hierarchical manner. For high-quality image data with a quality grade of 4 or above, it is written to the solid-state hard disk array for RAID5 storage. This storage method can provide a high-reliability and high-performance storage solution. RAID5 has a redundant data protection mechanism that can ensure that data is not lost when the hard disk fails, and provides good read and write performance, which is suitable for storing high-quality and important image data. These high-quality images are stored in the main storage area for further analysis, processing or archiving. For image data with quality levels between level 2 and level 3, they are written to the DDR4 memory for temporary storage to form a cache area. As a high-speed storage medium, DDR4 memory has low access latency and high transmission rate, which effectively supports temporary storage and fast access to these medium-quality image data. The storage in the cache area can accelerate the processing and analysis of image data, so that when performing real-time operations, the system can quickly obtain the required image data from the memory without waiting for the hard disk to read the operation. For low-quality image data with a quality level below level 1, they are written to the temporary buffer for short-term storage. Since these images are of poor quality and are not suitable for long-term storage or archiving, they are temporarily saved in a short-term storage area for subsequent processing or discarding. The high-quality image data stored in the main storage area is transmitted to the cloud storage server for remote archiving to ensure the long-term preservation of the image data, and enable the data to be remotely accessed and managed through the Internet to obtain distributed storage of image data.

[0023] The image data obtained from the distributed storage is input into the spatial guidance module to extract features from the image data. In the spatial guidance module, the image data is processed by a set of 3×3 convolutional layers, which is responsible for extracting local features of the image, such as low-level visual information such as edges and textures. The image data is processed by a 7×7 dilated convolutional layer, where the dilation rate is set to 2. The dilated convolution introduces the dilation rate to expand the receptive field while reducing the amount of calculation, effectively capturing a wider range of contextual information in the image. The spatial guidance module can extract local features and effectively encode the global information of the image, and finally generate spatial feature data. Channel attention calculation is performed on the spatial feature data, and weighted processing is performed according to the feature importance of each channel. In the channel attention mechanism, the channel weight coefficient of each channel is obtained by weighting each channel in the spatial feature data. These weight coefficients represent the contribution of each channel in the image features. According to the weight coefficients, the spatial feature data is weighted to highlight important features and suppress unimportant features to obtain weighted spatial features. At the same time, semantic features are extracted from the image data. The distributed stored image data is input into the deep convolutional network, and the deep convolutional network is used to extract features of the image layer by layer to obtain the semantic feature data of the image. The deep convolutional network extracts higher-level semantic information from the image through multi-layer convolution operations, such as object categories and scene structures. The obtained semantic feature data has a high dimension, so it is subjected to dimensionality reduction. Through dimensionality reduction operations, the semantic feature data is compressed to 128 channels. The weighted spatial features and semantic feature data are input into four groups of parallel pooling branches in the pooling feature enhancement module for multi-scale feature extraction. Each group of parallel pooling branches uses pooling kernels of different sizes, including pooling kernels of different scales such as 1×3, 3×1, 3×3 and 5×5. Through multi-scale pooling operations, information of different scales in the image is captured, and the diversity and expression ability of image features are improved. Each pooling branch performs effective feature extraction for different regions and scales of the image, and the obtained multi-scale feature data contains rich image information. On this basis, strip pooling kernels are used for long-distance dependency modeling. The sizes of the strip pooling kernels are 1×W and H×1, respectively, which can extract features in the long direction of the image and capture long-distance dependencies in the image. Through strip pooling, the global information modeling of the image is enhanced, the association between long-distance pixels in the image is captured, and the semantic understanding ability of the image is improved. The obtained strip pooling features effectively supplement the long-distance contextual information in the image. The strip pooling features and multi-scale feature data are weighted and summed to obtain the fused feature map. The fused feature map contains information from different scales, different spatial positions, and different semantic levels, and is a comprehensive feature representation of the image. Through a three-layer deconvolution network, the fused feature map is restored to the resolution of the original image.The deconvolution operation gradually restores the spatial dimensions of the feature map, so that the final restored image maintains the integrity of the feature information while restoring the same size and details as the original image. The deconvolution operation refines the spatial structure of the image, thereby improving the visual quality and detail performance of the image.

[0024] In the embodiment of the present invention, by designing an independent signal conditioning circuit and an adaptive gain control algorithm, high-precision acquisition and dynamic range adjustment of multi-channel image signals are achieved, and the quality and stability of image acquisition are effectively improved. Multi-level digital filtering and image enhancement processing are adopted, combined with feature extraction technology, to significantly improve the signal-to-noise ratio and clarity of the image, and a multi-dimensional image quality evaluation mechanism is innovatively introduced. Through the comprehensive evaluation of the three dimensions of clarity, noise and dynamic range, the accurate quantification of image quality is achieved. A hierarchical storage strategy based on quality level is designed, and the storage efficiency of the system is optimized by reasonably allocating storage resources, while ensuring the security and accessibility of important data. The dual-branch feature extraction network and pooling feature enhancement module are adopted to effectively retain the spatial information and semantic information of the image, and improve the accuracy and robustness of feature extraction. By integrating the FPGA processing platform and the multi-level storage structure, efficient utilization of system resources is achieved, and the real-time processing capability and data management efficiency of the system are significantly improved.

[0025] In a specific embodiment, the process of executing step S100 may specifically include the following steps: Amplifying the multi-channel image signals through a programmable gain amplifier circuit to obtain an amplified image signal, and performing low-pass filtering on the amplified image signal through a second-order Butterworth filter circuit to obtain a filtered image signal; The filtered image signal is input into a 14-bit high-speed AD conversion chip for analog-to-digital conversion to obtain a digitized signal, and the digitized signal is clock synchronized through a phase-locked loop circuit to obtain a multi-channel synchronized digitized signal; Input the multi-channel synchronous digitized signals into the data cache module inside the FPGA for cache processing to obtain the cached digitized signals, wherein the data cache module adopts a dual-port RAM structure; The data format of the buffered digitized signal is converted to obtain digitized image data.

[0026] Specifically, the multi-channel image signals are amplified and filtered to ensure the accuracy and signal quality of subsequent processing. In the system, the gain of the input multi-channel image signals is adjusted through the programmable gain amplifier circuit. The multi-channel image signals are analog signals transmitted through different channels and have different amplitudes and noise. Through the programmable gain amplifier circuit, the gain of each signal channel is adjusted according to actual needs to ensure that the amplitude of the image signal is suitable for subsequent processing and avoid distortion or saturation caused by too weak or too strong signals. Assume that the input image signal is , and its output signal after gain adjustment is The gain factor of the gain amplifier circuit is , then the image signal after gain is: ; in, is the gain factor of the programmable gain amplifier circuit, and are the time domain representations of the input signal and the output signal respectively. The signal after gain amplification contains a certain amount of high-frequency noise components. Next, the amplified image signal is low-pass filtered through a second-order Butterworth filter circuit. The Butterworth filter is a filter with a smooth frequency response that can effectively suppress high-frequency noise and retain the low-frequency part of the signal. The transfer function of the second-order Butterworth filter It is expressed as: ; in, It is a complex frequency domain variable. The characteristics of the second-order system determine that it has a flat passband and good cutoff characteristics. After being processed by the filter, the frequency components of the signal are smoothed, and the filtered image signal The image signal after filtering is input into a 14-bit high-speed AD conversion chip for analog-to-digital conversion, which converts the continuous analog signal into a discrete digital signal for digital circuit processing. Converted into digital signal , and the sampling frequency of the AD converter is , then the converted digital signal The relationship with the analog signal is expressed by the following formula: ; in, is the reference voltage, is the maximum digital value that can be represented by 14-bit precision, and round represents rounding operation. This conversion quantizes the amplitude of the analog signal into a 14-bit digital signal. After completing the analog-to-digital conversion, the digital signal is clock synchronized to ensure that the image data collected by multiple channels has consistent timing information. A phase-locked loop (PLL) circuit is used to synchronize the digital signal. The phase-locked loop circuit adjusts the phase and frequency of the digital signal so that the signals of multiple channels can be synchronized at the same time to avoid timing confusion. Through the clock synchronization of the phase-locked loop circuit, the multi-channel synchronized digitized signal obtained is expressed as: ; in, It is an unsynchronized digital signal. is the phase information adjusted by the phase-locked loop, It is a digital signal after clock synchronization. The multi-channel digital signal after clock synchronization is input into the data cache module inside the FPGA for cache processing. In FPGA, the data cache module adopts a dual-port RAM structure, that is, each data unit has two independent ports, one for writing data and the other for reading data. The dual-port RAM structure can achieve high-speed data access and allow simultaneous read and write operations within one clock cycle, thereby improving data processing efficiency. The cached digital signal is expressed as: ; in, It is a digital signal after clock synchronization. is the cached signal. In order to facilitate subsequent image processing and analysis, the cached digital signal is converted to a data format. The format of the image data needs to meet specific standards, such as RGB, YUV, etc., for further processing. The process of data format conversion includes converting the cached digital signal to an appropriate color space and resolution. Assume that the cached signal is , the target format is RGB, after format conversion, the obtained digital image data It is expressed as: ; in, It is the converted digital image data, and convert indicates the format conversion operation. The conversion process includes the conversion of color space (for example, from YCbCr to RGB) and the mapping of pixel values ​​(for example, from digital values ​​to image color values).

[0027] In a specific embodiment, the process of executing step S200 may specifically include the following steps: Perform histogram statistical analysis on the digitized image data to obtain the brightness value of each channel image, and compare and analyze the brightness value of each channel image with the preset target brightness range to obtain brightness difference data; The gain adjustment parameter is calculated by a proportional-integral control algorithm according to the brightness difference data to obtain a gain control amount, and the gain control amount is subjected to slope limiting processing to obtain a gain parameter after limiting; The limited gain parameter is written into the parameter storage area inside the FPGA to obtain gain parameter storage data, a gain lookup table is generated based on the gain parameter storage data, and gain correction is performed on the digitized image data to obtain image data with adjusted gain.

[0028] Specifically, a histogram statistical analysis is performed on the digitized image data. The image data consists of multiple channels (for example, RGB channels), and the signal of each channel represents the intensity of different colors. Assume that the image data is represented in matrix form, where is the red channel, It's the green channel. is the blue channel, and are the horizontal and vertical coordinates of the pixel respectively. For each channel, calculate the brightness value of the channel image, and get the brightness value by weighted average of the color value of each pixel. Assume that the brightness value of the image is expressed as , the image brightness value is calculated by the following formula: ; Among them, 0.299, 0.587, and 0.114 are common RGB to brightness weighting coefficients. The brightness value of each channel is counted to generate the brightness histogram of the channel and analyze the brightness distribution of the image. This process can reveal the overall brightness characteristics of the image and the existence of overexposed or underexposed areas. After obtaining the brightness histogram of each channel, the brightness value of each channel is compared and analyzed with the preset target brightness range. Assume that the target brightness range is arrive ,in and are the minimum and maximum brightness values ​​of the target image, respectively. The brightness difference is calculated to evaluate whether the brightness of the image meets the target range. ( ), brightness difference Calculated by the following formula: ; in, It is a channel In pixels The brightness value of the position, is the median of the target brightness range, is the total number of pixels in the image, and The width and height of the image respectively. Brightness difference data Reflects the degree of deviation between the actual brightness of the image and the target brightness range. The proportional-integral control algorithm (PI control algorithm) is used to calculate the gain adjustment parameter. The PI control algorithm generates a gain adjustment amount by calculating the brightness difference, and adjusts the gain of the image signal according to the adjustment amount to adjust the image brightness to the target range. The output gain control amount of the PI controller It is expressed as: ; in, is the proportional gain coefficient, is the integral gain coefficient, is the current brightness difference, is the integral term of the brightness difference. The gain is adjusted according to the current brightness deviation, and the integral part This takes into account historical brightness deviations and helps eliminate ongoing system deviations. It is a key parameter for adjusting the image brightness. In order to prevent instability or distortion caused by excessive or insufficient gain, the gain control amount is slope-limited, that is, the gain change rate is limited to avoid drastic fluctuations. The gain parameter after limitation It is expressed as: ; in, and are the minimum and maximum values ​​of the gain, respectively, and the function Indicates that Restricted to If the gain exceeds this range, it is limited to the maximum or minimum value. Write to the parameter storage area inside the FPGA. This storage area is used to store the gain parameters of each channel for subsequent gain calibration. Assume that the gain parameter storage data in the storage area is , then the data is expressed as: ; in, , , The limiting gain parameters of the red, green and blue channels are respectively. Based on these stored data, a gain lookup table (LUT) is generated for subsequent gain correction. Gain Lookup Table is a mapping table containing the gain adjustment value corresponding to each input brightness value, which is used to dynamically adjust the image brightness during gain correction. For gain correction of digitized image data, assume that the corrected image data is The gain correction process is achieved by matching the brightness value of each pixel with the gain lookup table. The correction formula for image data is: ; in, It's pixels The brightness value, is the original image data, is the gain adjustment factor corresponding to the brightness value. In this way, each channel of the image is adjusted according to the calculated gain, so that the brightness of the entire image meets the expected target range, completing the gain correction.

[0029] Among them, the gain adjustment parameter is calculated according to the brightness difference data through the proportional-integral control algorithm to obtain the gain control amount, including: initializing the population through the improved Sine chaotic map to obtain the initial gain parameter distribution, wherein the iterative expression of the Sine chaotic map is Xn+1=sin(πXn)×β, β is the chaos factor; the linear weight coefficient of the initial gain parameter distribution is calculated to obtain the weight coefficient matrix, wherein the linear weight coefficient adopts a time-varying decreasing strategy, the initial value is 0.9, and the termination value is 0.4; based on the weight coefficient matrix, the adaptive acceleration factor is calculated to obtain the acceleration factor sequence, wherein the adaptive acceleration factor is inversely proportional to the Euclidean distance between the individual current position and the global optimal position; the nonlinear convergence factor is constructed for the acceleration factor sequence to obtain the convergence control parameter, wherein the nonlinear convergence factor adopts an exponential decay model , the attenuation coefficient is logarithmically related to the number of iterations; based on the convergence control parameter, the water wave factor is modulated on the optimal position of the group to obtain the modulated position parameter, wherein the water wave factor is dynamically adjusted with the individual fitness value, and the adjustment range is [0.1, 0.9]; the modulated position parameter is updated through the sparrow elite mutation strategy to obtain the mutated position parameter, wherein the individuals with the top 20% of the fitness value are selected as the elite population; the mutated position parameter is input into the Bernoulli chaotic map for perturbation processing to obtain the perturbed position parameter, wherein the threshold of the Bernoulli chaotic map is 0.5; the gain control amount is optimized and calculated according to the perturbed position parameter to obtain the optimal gain control amount, wherein the exponential fitness function is used to evaluate the quality of the gain parameter, and the optimal gain control amount is used for the slope limiting step.

[0030] In a specific embodiment, the process of executing step S300 may specifically include the following steps: Inputting the image data with adjusted gain into a fifth-order FIR filter for digital filtering to obtain filtered image data, and performing gamma correction on the filtered image data through a lookup table to obtain gamma-corrected image data; Performing edge detection on the gamma-corrected image data by using a Sobel operator to obtain edge detection data, and performing edge sharpening on the gamma-corrected image data according to the edge detection data to obtain edge-enhanced image data; Feature detection is performed on the edge-enhanced image data to obtain feature point data, and the feature point data and the edge-enhanced image data are transmitted through the FPGA internal data bus to obtain processed image data.

[0031] Specifically, the image data with adjusted gain is input into a fifth-order FIR filter for digital filtering. The FIR filter is a digital filter that removes high-frequency noise and other unnecessary signals by weighted averaging the input signal. The filtering formula of the fifth-order FIR filter is expressed as: ; in, is the filtered image data, is the input image signal, is the filter coefficient of the FIR filter, is the current moment. The filter convolves the input signal with these coefficients to obtain a smooth output signal. The order of the fifth-order FIR filter determines the degree of smoothing of the signal. After this processing, the filtered image data can effectively remove noise and unnecessary high-frequency components, making the image clearer. Gamma correction is performed on the filtered image data. Gamma correction is a nonlinear transformation technology used in image processing, the purpose of which is to adjust the brightness or contrast of the image to adapt to the characteristics of the display device or the visual perception of the human eye. Gamma correction is implemented through a lookup table (LUT), in which each input pixel value is adjusted according to a preset gamma value. Assume that the pixel value of the input image is , pixel value after gamma correction Calculated by the following formula: ; in, is the original pixel value of the input image, is the gamma value, It is the pixel value after gamma correction, and 255 is the maximum pixel value. Controls the degree of brightness adjustment. will make the image darker, and The image becomes brighter. After gamma correction, the brightness distribution of the image is more consistent with the perception characteristics of the human eye, which helps to improve the visualization effect of the image. The Sobel operator is used to detect edges on the gamma-corrected image data. The Sobel operator is an edge detection algorithm that identifies the edges in the image by calculating the gradient of each pixel in the image. The Sobel operator is actually a convolution operator that uses two 3×3 convolution kernels to detect the gradient in the horizontal and vertical directions respectively. Assume that the image is at position The pixel value is , the gradient calculated by the Sobel operator is expressed as: ; ; in, and is the horizontal and vertical convolution kernel of the Sobel operator, defined as: ; By calculation and , get the gradient amplitude of each pixel in the image and the gradient direction , the formula is as follows: ; Image data after edge detection It reflects the strength of the edge in the image and can help extract important structures and features in the image. After edge detection, the edge of the gamma-corrected image is sharpened based on the edge detection data. Edge sharpening is a method of image enhancement that enhances the clarity of the image by highlighting the edge information in the image. Edge sharpening is achieved by weighted combination of the image and the edge detection result. Assume that the image after edge sharpening is , and its calculation formula is expressed as: ; in, is a weighting coefficient that controls the strength of edge enhancement. is the edge intensity map obtained by the Sobel operator. In this way, the details and edge information of the image are enhanced, making the image clearer. Feature detection is performed on the edge-enhanced image data. Feature detection is a method for extracting key points and key features in an image. These features are used for subsequent tasks such as image matching, object recognition, or image registration. Feature detection algorithms include Harris corner detection, SIFT (scale-invariant feature transform), and SURF (speeded up robust features). Assume that the feature points extracted by the feature detection algorithm are ,in It is The location of the feature point. The feature point data and the edge-enhanced image data are transmitted through the data bus inside the FPGA to achieve subsequent data processing or storage. The data bus inside the FPGA transmits the image data and feature point data to the next processing unit through a high-speed transmission interface, or stores it in the cache for further processing. The transmitted image data and feature point data Together they constitute the complete processed image data.

[0032] In this embodiment, before performing adaptive gain control on the digitized image data and performing digital filtering on the image data with adjusted gain, the method further includes: performing multi-scale decomposition processing on the image data with adjusted gain by using a complementary set empirical mode decomposition algorithm to obtain image component data of different frequency scales; performing stationarity analysis processing on the image component data of different frequency scales by using an augmented Dickey-Fuller test to obtain stationary component data and non-stationary component data; performing autocorrelation analysis processing on the stationary component data to obtain an autocorrelation coefficient sequence, and determining the order parameter of the autoregressive sliding average model based on the autocorrelation coefficient sequence; establishing an autoregressive sliding average prediction model for the stationary component data based on the order parameter to obtain a stationary component prediction result; performing sample entropy calculation processing on the non-stationary component data to obtain an entropy value feature vector, and determining the structural parameters of the radial basis neural network based on the entropy value feature vector; establishing an entropy-radial basis neural network prediction model for the non-stationary component data based on the structural parameters to obtain a non-stationary component prediction result; performing weighted fusion processing on the stationary component prediction result and the non-stationary component prediction result to obtain image component fusion data; performing signal reconstruction processing on the image component fusion data to obtain reconstructed image data, and the reconstructed image data is used for the step of performing digital filtering.

[0033] In a specific embodiment, the process of executing step S400 may specifically include the following steps: Perform energy gradient calculation on the processed image data to obtain the image average gradient value and gradient variance data, and compare the gradient variance data with the standard test image to obtain the clarity evaluation result; The signal-to-noise ratio of the processed image data is calculated by local mean filtering to obtain signal-to-noise ratio data, and the signal-to-noise ratio data is compared with a preset threshold to obtain a noise evaluation result; Performing histogram analysis on the processed image data to obtain histogram distribution data, and comparing the effective pixel coverage in the histogram distribution data with the digital range to obtain a dynamic range evaluation result; Generate a 4-bit status code based on the clarity evaluation result to obtain high-bit status data, wherein the 4-bit status code represents the image quality level; A 4-bit status code is generated based on the noise evaluation result and the dynamic range evaluation result to obtain low-bit status data, and the high-bit status data is combined with the low-bit status data to obtain an image quality evaluation status word.

[0034] Specifically, the energy gradient of the processed image data is calculated to obtain the average gradient value and gradient variance data of the image. The gradient of the image can reflect the rate of change of the pixel value in the image and is used to evaluate the clarity of the image. The basic principle of gradient calculation is to calculate the gradient vector of each pixel point, which represents the change of the pixel value in the horizontal and vertical directions. Assume that the original pixel value of the image is , the gradient calculation is performed using the Sobel operator or a higher-order operator. The common gradient calculation formula is: ; ; in, and They are the horizontal and vertical convolution kernels of the Sobel operator, is the pixel value in the neighborhood in the image, and is the gradient component in the horizontal and vertical directions. The gradient amplitude is calculated for: ; Calculate the average gradient of an image and the gradient variance : ; ; in, is the total number of pixels in the image, and are the width and height of the image respectively. The gradient variance reflects the distribution of the image edge. The larger the variance, the clearer the edge in the image and the higher the clarity. The clarity evaluation result is obtained by comparing with the standard test image. If the gradient variance is greater than a certain standard value, the image clarity is considered to be high, otherwise it is considered to be low. The signal-to-noise ratio (SNR) is calculated using local mean filtering. The signal-to-noise ratio is an indicator of the image noise level, which indicates the relative strength of the image signal and noise. The calculation of the signal-to-noise ratio is based on the mean and variance of the image, and the formula is: ; in, is the mean of the image, is the variance of the noise. Local mean filtering is used to extract the signal part of the image. By performing local window processing on the image, the noise level of the image can be effectively estimated. The signal-to-noise ratio data is compared with the preset threshold. If the signal-to-noise ratio is greater than a certain threshold, it is considered that there is less noise and the image quality is good. Histogram analysis is performed on the image data to evaluate the dynamic range of the image. The dynamic range reflects the range of brightness levels that the image can represent and determines the image's ability to retain details. The histogram of the image represents the distribution of pixel values, and the dynamic range is calculated using the following formula: ; in, and are the maximum and minimum pixel values ​​in the image, respectively. is 255. The dynamic range evaluation result is obtained by comparing the effective pixel coverage of the histogram with the digital range. If the coverage of the effective pixels is close to the maximum value of the digital range, it indicates that the image has a higher dynamic range and can display more details. After obtaining the clarity evaluation, noise evaluation and dynamic range evaluation results, a status code for the image quality is generated. A 4-bit status code is generated based on the clarity evaluation result to indicate the quality level of the image. Assume that the result of the clarity evaluation is compared with the standard by calculating the gradient variance. If the image is clear, a status code 1111 (indicating the highest image quality) is generated. If it is not clear, a corresponding lower status code is generated. Another 4-bit status code is generated based on the noise evaluation result and the dynamic range evaluation result to reflect the noise and dynamic range levels of the image. The high-order status code of the clarity evaluation result and the low-order status codes of the noise and dynamic range evaluation results are combined to form an image quality evaluation status word. Assume that the high-order status code is , the low status code is , then the final image quality assessment status word for: ; Among them, <<4 means that the high-order status code is shifted left by 4 bits, and then a bitwise OR operation is performed with the low-order status code to obtain the final 8-bit image quality assessment status word.

[0035] In a specific embodiment, the process of executing step S500 may specifically include the following steps: Parsing the image quality assessment status word to obtain image quality grade data, wherein the quality grade is determined according to the upper 4 bits of the image quality assessment status word; Generate a data index table for the processed image data to obtain image index data, wherein the data index table includes image timestamp, quality level and storage location information; Writing the processed image data with a quality level of 4 or above into a solid state drive array for RAID5 storage to obtain image data in a primary storage area, and writing the processed image data with a quality level between 2 and 3 into a DDR4 memory for temporary storage to obtain image data in a cache area; The processed image data with a quality level below level 1 is written into a temporary buffer for short-term storage to obtain image data in a temporary storage area, and the image data in the main storage area is transmitted to a cloud storage server for remote archiving to obtain distributed storage image data.

[0036] Specifically, the image quality assessment status word is parsed, and the quality level of the image is determined based on the upper 4 bits of the status word. Assume that the image quality assessment status word is an 8-bit binary number, in which the upper 4 bits represent the composite assessment result of clarity and noise level, and the lower 4 bits represent the data of dynamic range and other image features. The quality level of the image can be directly obtained by extracting the upper 4 bits from this status word. For example, if the quality assessment status word is 11011001, 1101 represents the quality level, and 1001 represents other assessment features. Convert the upper 4 bits 1101 to a decimal number, and the quality level is 13, indicating that the image quality is at a high level. If the upper 4 bits are 0000, it means that the quality level is the lowest. The quality level can be divided into multiple levels, such as 1 to 4, where the higher the level, the better the image quality. Generate a data index table for subsequent storage and management of image data. The data index table is used to store key information of the image, including timestamp, quality level, and storage location information. The timestamp is used to record the acquisition time of the image, the quality level comes from the previous image quality assessment, and the storage location information is used to indicate the specific location of the image data in the storage medium. Assume that the processed image data is saved as multiple files or data blocks, each file or data block has a unique identifier, and these data blocks are stored in different storage media. After the data index table is generated, different storage media are selected according to the quality level of the image. Image data with an image quality level of 4 or above represents high-quality images, and these images need to be stored first. In order to ensure the storage speed and reliability of high-quality images, these image data are stored in a solid-state drive array (RAID5). RAID5 is a redundant array technology that ensures data security through distributed storage and improves access speed. Assuming that the image quality level is 4, it is stored in the RAID5 array to obtain the image data in the main storage area. For images with quality levels between 2 and 3, these images have moderate quality and are temporarily stored in a more economical way. DDR4 memory has a high read and write speed and is suitable for short-term storage and fast access. Image data with quality levels of 2 to 3 are stored in the DDR4 memory in the cache area. This storage method ensures efficient access to images in the short term, and the memory capacity is usually sufficient to meet regular image data storage needs. Images with a quality level below level 1 do not need to be retained for a long time due to their poor quality. In order to save storage space and resources, these images are temporarily stored in low-speed storage media, such as temporary buffers. These image data will be deleted when the system is idle, or overwritten under certain conditions. The function of the temporary buffer is to store image data with lower quality that does not need to be stored for a long time. After the high-quality image data in the main storage area is processed, the system regularly uploads these image data to the cloud storage server for remote archiving to obtain distributed storage image data.The operation of uploading image data to the cloud storage server usually adopts batch processing mode, and the data stored in the RAID5 array is transmitted to the remote server through the network for off-site backup. The upload process needs to ensure the integrity of the data and the security of the transmission, and encryption and verification mechanisms are used to ensure the security of the image data.

[0037] In a specific embodiment, executing the image acquisition method based on the multi-channel image acquisition card also includes the following steps: The distributed stored image data is input into the spatial guidance module, and feature extraction is performed through a 3×3 convolution layer and a 7×7 dilated convolution layer to obtain spatial feature data, where the dilated rate of the dilated convolution layer is 2; Perform channel attention calculation on the spatial feature data to obtain the channel weight coefficient, and perform weighted processing on the spatial feature data according to the channel weight coefficient to obtain the weighted spatial feature; The distributed stored image data is input into the deep convolutional network for semantic feature extraction to obtain semantic feature data, and the semantic feature data is reduced in dimension to obtain 128-channel semantic features; The weighted spatial features and semantic features are respectively input into the four groups of parallel pooling branches in the pooling feature enhancement module for multi-scale feature extraction to obtain multi-scale feature data, where the pooling kernel sizes of the four groups of parallel pooling branches are 1×3, 3×1, 3×3, and 5×5 respectively; The multi-scale feature data is modeled with long-distance dependencies through strip pooling kernels to obtain strip pooling features, where the sizes of the strip pooling kernels are 1×W and H×1; The strip pooling features and multi-scale feature data are weightedly summed to obtain a fused feature map, which is then restored to the original image resolution through a three-layer deconvolution network.

[0038] Specifically, the distributed stored image data is input into the spatial guidance module for feature extraction. In the spatial guidance module, a The convolution layer extracts local features. This convolution layer performs preliminary local feature extraction on the image and captures the low-level features of the image (such as edges, textures, etc.). , the output feature map of the convolution operation It is expressed by the following formula: ; in, yes The weight of the convolution kernel, is the pixel value of a point in the image data, is the output feature map after convolution. During the convolution process, convolution operation is performed on each pixel position to extract the spatial features of the image. The dilated convolution layer performs feature extraction. The dilated convolution can increase the receptive field while keeping the computational complexity relatively low, thereby extracting richer spatial information. In the dilated convolution, the dilation rate is 2, that is, the interval between the convolution kernels is 2, which enables the convolution operation to span a larger area, thereby capturing a wider range of structural information in the image. The dilated convolution operation is expressed by the following formula: ; in, yes The convolution kernel has a hole rate of 2, that is, there is a gap of 2 between each convolution kernel element. In this way, the hole convolution can increase the receptive field and extract a wider range of spatial features. After feature extraction, channel attention calculation is performed. The channel attention mechanism can dynamically adjust the weight of the feature according to the contribution of each channel to the final output. By calculating the weight coefficient of each channel , weighted processing is performed on each channel to increase the model's attention to important channels. The calculation of channel attention uses a combination of global average pooling and a fully connected layer to compress the feature vector of each channel into a scalar weight. The calculation formula is as follows: ; in, Indicates The characteristics of the channels, GlobalAvgPool Indicates global average pooling of each channel feature map. MLP is a multi-layer perceptron used to generate channel weights. is an activation function (such as Sigmoid). By weighting the features of each channel, we get the weighted spatial features. : ; The weighted spatial features and the input image data are fed into a deep convolutional network for semantic feature extraction. The deep convolutional network learns higher-level abstract features from the image, such as semantic information such as objects and scenes. When extracting semantic features, the convolutional layer gradually extracts complex patterns and semantic information in the image. The obtained semantic features The semantic information of different regions in the image is represented by similar convolution operations. In order to improve the expressive power of the features, the semantic features are reduced in dimension and compressed to 128 channels. This step is completed by a dimensionality reduction convolution layer, and the 128 channels output by it can effectively represent the high-level semantic information in the image. The dimensionality reduction operation is implemented by the following formula: ; is the semantic feature after dimensionality reduction, with a channel number of 128. After spatial feature weighting and semantic feature extraction, all features are sent to the pooling feature enhancement module. This module contains four sets of parallel pooling branches, which are used to extract multi-scale features of images at different scales. The pooling kernel sizes of these pooling branches are , , ,and Each pooling kernel can capture the feature information of the image at different scales. The result of the pooling operation is multiple feature maps. Through the pooling operation, multi-scale feature data is obtained. , which expresses the features of the image at different scales. After multi-scale feature extraction, long-distance dependency modeling is performed through strip pooling kernels. Strip pooling kernels can effectively capture the association between long-distance pixels in the image, which helps to process long-distance dependencies and maintain the global information of the image. The size of the strip pooling kernel is and ,in and are the width and height of the image. The strip pooling operation is expressed as follows: ; In this way, pooled features with global long-distance dependencies are obtained . Perform weighted summation on the strip pooling features and multi-scale feature data to obtain the fused feature map The weighted sum operation is expressed as: ; in, and is a weighting coefficient that controls the fusion ratio of different features. The fused feature map is input into the three-layer deconvolution network for image restoration. The deconvolution network restores the fused feature map to the original image resolution through layer-by-layer upsampling operations. The restoration process is expressed by the following formula: ; in, It is the restored image. The deconvolution operation gradually enlarges the fused feature map until it is restored to the same resolution as the original image.

[0039] See also Figure 2 , Figure 2 The structure schematic block diagram of the image acquisition device 200 based on the multi-channel image acquisition card provided in the embodiment of the present application is as follows: Figure 2 As shown, the image acquisition device 200 based on the multi-channel image acquisition card includes: The signal conditioning module 210 is used to input the multi-channel image signals into the multi-channel image acquisition card for signal conditioning to obtain digital image data; A gain control module 220, configured to perform adaptive gain control on the digitized image data to obtain image data with adjusted gain; The feature extraction module 230 is used to perform digital filtering, image enhancement and feature extraction on the image data with adjusted gain to obtain processed image data; An evaluation module 240 is used to perform clarity evaluation, noise evaluation and dynamic range evaluation on the processed image data to obtain an image quality evaluation status word; The hierarchical storage module 250 is used to perform hierarchical storage on the processed image data according to the image quality evaluation status word to obtain distributed stored image data.

[0040] Through the collaboration of the above components, by designing independent signal conditioning circuits and adaptive gain control algorithms, high-precision acquisition and dynamic range adjustment of multi-channel image signals are achieved, effectively improving the quality and stability of image acquisition. Multi-level digital filtering and image enhancement processing are adopted, combined with feature extraction technology, which significantly improves the signal-to-noise ratio and clarity of the image. The multi-dimensional image quality evaluation mechanism is innovatively introduced, and the image quality is accurately quantified through comprehensive evaluation of the three dimensions of clarity, noise and dynamic range. A hierarchical storage strategy based on quality level is designed, which optimizes the storage efficiency of the system by reasonably allocating storage resources, while ensuring the security and accessibility of important data. The dual-branch feature extraction network and pooling feature enhancement module are adopted to effectively retain the spatial information and semantic information of the image, and improve the accuracy and robustness of feature extraction. By integrating the FPGA processing platform and multi-level storage structure, efficient utilization of system resources is achieved, and the real-time processing capability and data management efficiency of the system are significantly improved.

[0041] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0042] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0043] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An image acquisition method based on a multi-channel image acquisition card, characterized in that: include: Inputting multiple image signals into a multi-channel image acquisition card for signal conditioning to obtain digital image data; Performing adaptive gain control on the digitized image data to obtain image data with adjusted gain; Performing digital filtering, image enhancement and feature extraction on the image data whose gain has been adjusted to obtain processed image data; Performing clarity assessment, noise assessment and dynamic range assessment on the processed image data to obtain an image quality assessment status word; The processed image data is hierarchically stored according to the image quality evaluation status word to obtain distributedly stored image data.

2. The image acquisition method based on a multi-channel image acquisition card according to claim 1, characterized in that: The method of inputting the multi-channel image signal into the multi-channel image acquisition card for signal conditioning to obtain digital image data includes: Amplifying the multi-channel image signals through a programmable gain amplifier circuit to obtain an amplified image signal, and performing low-pass filtering on the amplified image signal through a second-order Butterworth filter circuit to obtain a filtered image signal; Inputting the filtered image signal into a 14-bit high-speed AD conversion chip for analog-to-digital conversion to obtain a digitized signal, and performing clock synchronization on the digitized signal through a phase-locked loop circuit to obtain a multi-channel synchronized digitized signal; Inputting the multi-channel synchronized digitized signals into a data cache module inside the FPGA for cache processing to obtain cached digitized signals, wherein the data cache module adopts a dual-port RAM structure; The data format of the buffered digitized signal is converted to obtain digitized image data.

3. The image acquisition method based on a multi-channel image acquisition card according to claim 1, characterized in that: The step of performing adaptive gain control on the digitized image data to obtain image data with adjusted gain includes: Performing histogram statistical analysis on the digitized image data to obtain the brightness value of each channel image, and performing comparative analysis on the brightness value of each channel image with a preset target brightness range to obtain brightness difference data; Calculating a gain adjustment parameter through a proportional-integral control algorithm according to the brightness difference data to obtain a gain control amount, and performing slope limiting processing on the gain control amount to obtain a limited gain parameter; The limited gain parameter is written into a parameter storage area inside the FPGA to obtain gain parameter storage data, a gain lookup table is generated based on the gain parameter storage data, and gain correction is performed on the digitized image data to obtain image data with adjusted gain.

4. The image acquisition method based on a multi-channel image acquisition card according to claim 1, characterized in that: The step of performing digital filtering, image enhancement and feature extraction on the image data with adjusted gain to obtain processed image data comprises: Inputting the image data with adjusted gain into a fifth-order FIR filter for digital filtering to obtain filtered image data, and performing gamma correction on the filtered image data through a lookup table to obtain gamma-corrected image data; Performing edge detection on the gamma-corrected image data by using a Sobel operator to obtain edge detection data, and performing edge sharpening on the gamma-corrected image data according to the edge detection data to obtain edge-enhanced image data; Feature detection is performed on the edge-enhanced image data to obtain feature point data, and the feature point data and the edge-enhanced image data are transmitted through an internal data bus of the FPGA to obtain processed image data.

5. The image acquisition method based on a multi-channel image acquisition card according to claim 1, characterized in that: The process of performing clarity assessment, noise assessment and dynamic range assessment on the processed image data to obtain an image quality assessment status word includes: Performing energy gradient calculation on the processed image data to obtain an image average gradient value and gradient variance data, and comparing the gradient variance data with a standard test image to obtain a clarity evaluation result; Calculating the signal-to-noise ratio of the processed image data by local mean filtering to obtain signal-to-noise ratio data, and comparing the signal-to-noise ratio data with a preset threshold to obtain a noise evaluation result; Performing a histogram analysis on the processed image data to obtain histogram distribution data, and comparing the effective pixel coverage in the histogram distribution data with the digital range to obtain a dynamic range evaluation result; Generate a 4-bit status code based on the clarity evaluation result to obtain high-bit status data, wherein the 4-bit status code represents the image quality level; A 4-bit status code is generated based on the noise evaluation result and the dynamic range evaluation result to obtain low-bit status data, and the high-bit status data is combined with the low-bit status data to obtain an image quality evaluation status word.

6. The image acquisition method based on a multi-channel image acquisition card according to claim 1, characterized in that: The step of performing hierarchical storage on the processed image data according to the image quality evaluation status word to obtain distributed stored image data includes: Parsing the image quality assessment status word to obtain image quality grade data, wherein the quality grade is determined according to the upper 4 bits of the image quality assessment status word; Generating a data index table for the processed image data to obtain image index data, wherein the data index table includes image timestamp, quality level and storage location information; Writing the processed image data with a quality level of 4 or above into a solid state drive array for RAID5 storage to obtain image data in a main storage area, and writing the processed image data with a quality level between 2 and 3 into a DDR4 memory for temporary storage to obtain image data in a cache area; The processed image data with a quality level below level 1 is written into a temporary buffer for short-term storage to obtain image data in a temporary storage area, and the image data in the main storage area is transmitted to a cloud storage server for remote archiving to obtain distributed storage image data.

7. The image acquisition method based on a multi-channel image acquisition card according to claim 1, characterized in that: The image acquisition method based on the multi-channel image acquisition card also includes: Input the distributed stored image data into a spatial guidance module, perform feature extraction through a 3×3 convolution layer and a 7×7 hole convolution layer, and obtain spatial feature data, wherein the hole rate of the hole convolution layer is 2; Performing channel attention calculation on the spatial feature data to obtain a channel weight coefficient, and performing weighted processing on the spatial feature data according to the channel weight coefficient to obtain a weighted spatial feature; Inputting the distributed stored image data into a deep convolutional network to extract semantic features to obtain semantic feature data, and performing dimension reduction on the semantic feature data to obtain 128-channel semantic features; Inputting the weighted spatial features and the semantic features into four groups of parallel pooling branches in a pooling feature enhancement module to extract multi-scale features, thereby obtaining multi-scale feature data, wherein the pooling kernel sizes of the four groups of parallel pooling branches are 1×3, 3×1, 3×3, and 5×5, respectively; Performing long-distance dependency modeling on the multi-scale feature data through a strip pooling kernel to obtain strip pooling features, wherein the sizes of the strip pooling kernel are 1×W and H×1; The strip pooling features and the multi-scale feature data are weightedly summed to obtain a fused feature map, and the fused feature map is restored to the original image resolution through a three-layer deconvolution network.

8. An image acquisition device based on a multi-channel image acquisition card, characterized in that: The method for performing an image acquisition method based on a multi-channel image acquisition card according to any one of claims 1 to 7 comprises: A signal conditioning module is used to input multiple image signals into a multi-channel image acquisition card for signal conditioning to obtain digital image data; A gain control module, used for performing adaptive gain control on the digitized image data to obtain image data with adjusted gain; A feature extraction module, used for performing digital filtering, image enhancement and feature extraction on the image data with adjusted gain to obtain processed image data; An evaluation module, used for performing clarity evaluation, noise evaluation and dynamic range evaluation on the processed image data to obtain an image quality evaluation status word; The hierarchical storage module is used to perform hierarchical storage on the processed image data according to the image quality evaluation status word to obtain distributed stored image data.

Images