Comparison and analysis method and system based on ai image recognition

Abstract

The present application relates to a comparison and analysis method and system based on AI image recognition. The system comprises: an image collection module, which comprises an image collection unit for collecting a craft image; an image processing module, which comprises a preprocessing unit for performing an operation on the craft image, so as to output a feature image; a comparison and analysis module, which comprises a comparison unit for comparing the craft image with a reference image, so as to output an image similarity degree; a light intensity adjustment module, which is used for determining a light intensity weight coefficient of an underexposed region on the basis of an average grayscale value of the craft image; a frequency adjustment module, which is used for determining a signal frequency hopping interval of the craft image on the basis of a transmission delay duration of the craft image; and a number adjustment module, which is used for determining a threshold value for the number of intra-frame prediction modes on the basis of an average encoding duration of several frames of craft images.

Description

A Comparison and Analysis Method and System Based on AI Image Recognition Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a comparison and analysis method and system based on AI image recognition. Background Technology

[0002] With the large-scale development of the handicrafts industry and the increasing demands of consumers for quality, traditional comparison and analysis methods relying on human experience are no longer sufficient to meet the industry's needs for efficiency, accuracy, and standardization, making technological upgrades an inevitable trend. With the rise of computer vision and artificial intelligence technologies, image recognition-based comparison and analysis technologies are gradually being applied to the handicrafts field. However, existing technical solutions still have many shortcomings and cannot fully meet practical needs.

[0003] In existing technologies, visualization methods based on AI image recognition typically include: S1, obtaining the actual type of the target object and dividing the target object into regions according to the actual type; S2, acquiring actual view information of each region through an AI imaging device; S3, optimizing and preprocessing the actual view information to obtain preprocessed view information; S4, extracting and processing the preprocessed view information to obtain actual image features; S5, comparing the preprocessed view information and the actual image features using AI image recognition technology to obtain a comparison result. It is evident that the aforementioned visualization methods, systems, terminals, and storage media based on AI image recognition suffer from a problem where, when the acquired image is underexposed, the image quality cannot be corrected through light intensity optimization, leading to distortion of the subsequently extracted actual image features, thus reducing the stability of AI image comparison analysis. Summary of the Invention

[0004] To address this issue, this application provides a comparison and analysis method and system based on AI image recognition, which overcomes the problem in the prior art where underexposure of the acquired image prevents the image quality from being corrected through light intensity optimization, leading to distortion of the actual image features extracted subsequently, and thus reducing the stability of AI image comparison analysis.

[0005] To achieve the above objectives, this application provides a comparison and analysis system based on AI image recognition, comprising:

[0006] The image acquisition module includes an image acquisition unit for acquiring images of handicrafts via a camera and an image transmission unit connected to the image acquisition unit for transmitting the images of handicrafts to a processing location;

[0007] An image processing module, connected to the image acquisition module, includes a preprocessing unit for sequentially performing denoising, cropping, encoding, and feature extraction operations on the handicraft image to output a feature image, and a model training unit connected to the preprocessing unit for training an initial model based on the feature image to output a machine learning model.

[0008] The comparison and analysis module, which is connected to the image processing module, includes a comparison unit for comparing the handicraft image with a reference image using the machine learning model to output image similarity, and an analysis unit connected to the comparison unit for analyzing the image similarity to output analysis results.

[0009] A light intensity adjustment module, which is connected to the image acquisition module, is used to determine the light intensity weighting coefficient of the underexposed area based on the average gray value of the handicraft image.

[0010] A frequency adjustment module, which is connected to the image acquisition module and the light intensity adjustment module respectively, is used to determine the signal frequency jump interval of the handicraft image based on the transmission delay time of the handicraft image;

[0011] The quantity adjustment module, which is connected to the image processing module and the frequency adjustment module respectively, is used to determine the quantity threshold of the intra-frame prediction mode based on the average encoding time of several frames of craft images.

[0012] Furthermore, the light intensity adjustment module responds to the fact that the average gray value of the craft image is greater than or equal to a preset second gray value, thus determining that the analytical stability of the AI ​​image comparison meets the requirements;

[0013] The light intensity adjustment module determines that the analysis stability of the AI ​​image comparison does not meet the requirements when the average gray value of the craft image is less than the preset second gray value.

[0014] Furthermore, in response to the average grayscale value of the craft image being greater than a preset first grayscale value and less than a preset second grayscale value, the light intensity adjustment module preliminarily determines that the signal integrity of the craft image does not meet the requirements.

[0015] Furthermore, the light intensity adjustment module increases the light intensity weighting coefficient of the underexposed area in response to the average gray value of the craft image being less than or equal to the preset first gray value;

[0016] The increase in the light intensity weighting coefficient of the underexposed area is determined by the difference between a preset first gray value and the average gray value of the craft image.

[0017] Furthermore, the frequency adjustment module responds to the fact that the transmission delay of the craft image is less than or equal to a preset first delay duration, and determines that the signal integrity of the craft image meets the requirements.

[0018] The frequency adjustment module determines that the signal integrity of the craft image does not meet the requirements when the transmission delay of the craft image is greater than the preset first delay duration.

[0019] Furthermore, the frequency adjustment module responds to the fact that the transmission delay of the craft image is greater than the preset first delay and less than or equal to the preset second delay by reducing the signal frequency jump interval of the craft image;

[0020] The frequency adjustment module responds to the fact that the transmission delay of the craft image is greater than the preset second delay duration, and preliminarily determines that the encoding accuracy of the craft image does not meet the requirements.

[0021] Furthermore, the reduction in the signal frequency switching interval of the craft image is determined by the difference between the transmission delay of the craft image and a preset first delay.

[0022] Furthermore, the quantity adjustment module determines that the encoding accuracy of the craft image meets the requirements when the average encoding duration of several frames of craft images is less than or equal to the preset encoding duration.

[0023] The quantity adjustment module responds to the fact that the average encoding duration of the several frames of craft images is greater than the preset encoding duration, determines that the encoding accuracy of the craft images does not meet the requirements, and reduces the quantity threshold of intra-frame prediction modes.

[0024] Furthermore, the reduction in the number threshold of the intra-frame prediction mode is determined by the difference between the average encoding time of several frames of craft images and the preset encoding time.

[0025] This application also provides a comparison and analysis method based on AI image recognition, including:

[0026] The image of the handicraft is captured by a camera and transmitted to the processing location. The handicraft image is then subjected to noise reduction, cropping, encoding and feature extraction operations to output a feature image. The initial model is trained based on the feature image to output a machine learning model.

[0027] The machine learning model is used to compare the image of the handicraft with a reference image to output image similarity, and the image similarity is analyzed to output analysis results;

[0028] The grayscale values ​​of several handicraft images are obtained, and the average grayscale value of the handicraft images is calculated. Based on the average grayscale value of the handicraft images, it is determined whether the analysis stability of the AI ​​image comparison meets the requirements.

[0029] If the analysis stability of the AI ​​image comparison does not meet the requirements, it is determined whether the light intensity weighting coefficient of the underexposed area needs to be increased.

[0030] If it is not necessary to increase the light intensity weighting coefficient of the underexposed area, then obtain the transmission delay time of the craft image to determine whether the signal integrity of the craft image meets the requirements.

[0031] If the signal integrity of the craft image does not meet the requirements, determine whether it is necessary to reduce the signal frequency transition interval of the craft image.

[0032] If it is not necessary to reduce the signal frequency jump interval of the craft images, the number threshold of intra-frame prediction modes is determined based on the average coding duration of several frames of craft images.

[0033] Compared with the prior art, the beneficial effects of this application are as follows: The system described in this application, by setting up an image acquisition module, an image processing module, a comparison and analysis module, a light intensity adjustment module, a frequency adjustment module, and a quantity adjustment module, adjusts the light intensity weighting coefficient of the underexposed area according to the average gray value of the craft image. Due to the slight displacement of the camera caused by the vibration of the workshop equipment, the angle of light received by the lens changes, resulting in a reduction in the amount of light entering the camera. If the parameters are not adjusted synchronously, underexposure will occur, thereby increasing the noise. By increasing the light intensity weighting coefficient of the underexposed area, the brightness of the underexposed area can be preferentially increased, accurately filling the light intensity gap caused by the reduction in the amount of light entering the camera, and avoiding overexposure of normal areas in order to compensate for underexposure. The signal frequency jump interval of the craft image is adjusted according to the transmission delay time of the craft image. Since multiple sets of equipment may share the same transmission network, signal superposition may occur. Interference can cause a decrease in transmission rate. By reducing the frequency transition interval of the image signal, the device can stay on a single frequency for a shorter time, reducing the probability of being hit by interference signals. This allows the system to quickly escape occupied channels, seize clean frequencies to transmit image data, reduce transmission interruptions caused by interference, and indirectly improve the effective transmission rate. The number threshold of intra-frame prediction modes is adjusted based on the average encoding time of several frames of craft images. Since the AI ​​image recognition system needs to process image acquisition, compression encoding, feature extraction, and comparison analysis simultaneously, if multiple tasks compete for the same hardware resources, it will lead to insufficient resources for the compression encoding component. By reducing the number threshold of intra-frame prediction modes, the amount of computation can be reduced, the resource requirements can be lowered, and the compression encoding component can work more efficiently with limited resources, reducing processing time and improving the stability of AI image comparison analysis. Attached Figure Description

[0034] Figure 1 is a block diagram of the overall structure of the AI ​​image recognition-based comparison and analysis system according to an embodiment of this application;

[0035] Figure 2 is an overall flowchart of the comparison and analysis method based on AI image recognition in an embodiment of this application;

[0036] Figure 3 is a flowchart of the process of determining the light intensity weighting coefficient of the underexposed area by the AI ​​image recognition-based comparison and analysis system in an embodiment of this application.

[0037] Figure 4 is a logic flowchart of the process of determining the signal frequency jump interval of the handicraft image by the AI ​​image recognition-based comparison and analysis system according to an embodiment of this application. Detailed Implementation

[0038] To make the objectives and advantages of this application clearer, the application will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining this application and are not intended to limit this application.

[0039] Preferred embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this application and are not intended to limit the scope of protection of this application.

[0040] Please refer to Figures 1, 2, 3, and 4, which are respectively the overall structural block diagram, overall flowchart, logical flowchart of the process for determining the light intensity weighting coefficient of the underexposed area, and logical flowchart of the process for determining the signal frequency jump interval of the craft image, based on the AI ​​image recognition comparison and analysis method and system of this application. This application provides an AI image recognition-based comparison and analysis system, including:

[0041] The image acquisition module includes an image acquisition unit for acquiring images of handicrafts via a camera and an image transmission unit connected to the image acquisition unit for transmitting the images of handicrafts to a processing location;

[0042] An image processing module, connected to the image acquisition module, includes a preprocessing unit for sequentially performing denoising, cropping, encoding, and feature extraction operations on the handicraft image to output a feature image, and a model training unit connected to the preprocessing unit for training an initial model based on the feature image to output a machine learning model.

[0043] The comparison and analysis module, which is connected to the image processing module, includes a comparison unit for comparing the handicraft image with a reference image using the machine learning model to output image similarity, and an analysis unit connected to the comparison unit for analyzing the image similarity to output analysis results.

[0044] A light intensity adjustment module, which is connected to the image acquisition module, is used to determine the light intensity weighting coefficient of the underexposed area based on the average gray value of the handicraft image.

[0045] A frequency adjustment module, which is connected to the image acquisition module and the light intensity adjustment module respectively, is used to determine the signal frequency jump interval of the handicraft image based on the transmission delay time of the handicraft image;

[0046] The quantity adjustment module, which is connected to the image processing module and the frequency adjustment module respectively, is used to determine the quantity threshold of the intra-frame prediction mode based on the average encoding time of several frames of craft images.

[0047] Specifically, images of handicrafts include the complete appearance of the handicraft, its decorative structure, and signs of wear on its surface.

[0048] Specifically, the feature images include the wear marks on the surface of the coded craftsmanship, the complete appearance of the denoised craftsmanship, and the decorative structure of the cut-out craftsmanship.

[0049] Specifically, machine learning models can be convolutional neural networks, Siamese neural networks, or support vector machines.

[0050] Specifically, the reference image is a high-quality image acquired through professional equipment and preprocessed and optimized (denoising, calibration, enhancement), containing the complete features of the artifact in an ideal or standard state.

[0051] Specifically, image similarity is a quantitative metric calculated using a machine learning model, used to describe the degree of matching between a craft image and a reference image in terms of visual features.

[0052] Specifically, the analysis results are the texture similarity between the craft image and the reference image, the difference region between the craft image and the reference image, and the contour similarity between the craft image and the reference image.

[0053] Specifically, the light intensity weighting coefficient of the underexposed region is a core parameter used to correct the interference of underexposure on feature extraction and similarity calculation. Its essence is to dynamically adjust the contribution of the underexposed region in the overall analysis, so as to avoid the non-substantial differences caused by insufficient light affecting the accuracy of the final comparison results.

[0054] Specifically, the signal frequency jump interval of the craft image is a key parameter for optimizing the stability and efficiency of image data transmission. It mainly targets the signal transmission process between the image acquisition module and the processing module. Its core meaning is the time interval between two adjacent signal frequency adjustments. By dynamically controlling this interval, adaptive optimization of the transmission link can be achieved.

[0055] Specifically, the threshold for the number of intra-prediction modes is a core control parameter set for the intra-prediction coding stage of an image. Its core meaning is the upper limit of the types of intra-prediction modes that the system allows to be used for the coding blocks (such as 4x4 and 8x8 pixel blocks) of a single frame of a craft image.

[0056] In implementation, the system described in this application, by setting up an image acquisition module, an image processing module, a comparison and analysis module, a light intensity adjustment module, a frequency adjustment module, and a quantity adjustment module, adjusts the light intensity weighting coefficient of underexposed areas based on the average grayscale value of the craft image. Due to slight camera displacement caused by vibrations from workshop equipment, the angle of light received by the lens changes, reducing the amount of light entering the camera. If the parameters are not adjusted synchronously, underexposure occurs, leading to increased noise. By increasing the light intensity weighting coefficient of underexposed areas, brightness enhancement can be prioritized for these areas, accurately filling the light intensity gap caused by reduced light intake and avoiding overexposure of normal areas in an attempt to compensate for underexposure. The system also adjusts the signal frequency jump interval of the craft image based on the transmission delay time. Since multiple devices may share the same transmission network, signal superposition interference may occur, leading to… The transmission rate decreases, but by reducing the signal frequency transition interval of the image, the device's dwell time on a single frequency is shortened, reducing the probability of being hit by interference signals. This allows the system to quickly escape occupied channels, seize clean frequencies to transmit image data, reduce transmission interruptions caused by interference, and indirectly improve the effective transmission rate. The threshold for the number of intra-frame prediction modes is adjusted based on the average encoding time of several frames of craft images. Since the AI ​​image recognition system needs to handle image acquisition, compression encoding, feature extraction, and comparison analysis simultaneously, if multiple tasks compete for the same hardware resources, it will lead to insufficient resources for the compression encoding component. By reducing the threshold for the number of intra-frame prediction modes, the computational load can be reduced, lowering the resource requirements. This allows the compression encoding component to work more efficiently with limited resources, reducing processing time and improving the stability of AI image comparison analysis.

[0057] Specifically, the light intensity adjustment module responds to the average gray value of the craft image being greater than or equal to a preset second gray value, thus determining that the analytical stability of the AI ​​image comparison meets the requirements;

[0058] The light intensity adjustment module determines that the analysis stability of the AI ​​image comparison does not meet the requirements when the average gray value of the craft image is less than the preset second gray value.

[0059] Specifically, the light intensity adjustment module responds to the fact that the average gray value of the craft image is greater than a preset first gray value and less than a preset second gray value, initially determines that the signal integrity of the craft image does not meet the requirements, and determines whether the signal integrity of the craft image meets the requirements based on the transmission delay time of the craft image.

[0060] It is understandable that the preset first grayscale value is less than the preset second grayscale value, and the three intervals divided by the preset first grayscale value and the preset second grayscale value correspond to three different situations:

[0061] The first interval is when the average gray value of the craft image is less than or equal to the preset first gray value. The corresponding situation is: due to the slight displacement of the camera caused by the vibration of the workshop equipment, the angle of light received by the lens changes, the amount of light decreases, and the parameters are not adjusted synchronously, resulting in underexposure, which in turn leads to increased noise.

[0062] The second interval is where the average gray value of the craft image is greater than the preset first gray value and less than the preset second gray value. The corresponding situation is that, since multiple sets of equipment may share the same transmission network, signal superposition interference may occur, which will lead to a decrease in transmission rate.

[0063] The third interval is when the average gray value of the craft image is greater than or equal to the preset second gray value, which corresponds to the situation where the analysis stability of the AI ​​image comparison meets the requirements.

[0064] Understandably, using the first and second grayscale values ​​to characterize the analytical stability of AI image comparison essentially involves quantifying the boundary values ​​of the overall brightness of the image to quickly filter out image quality ranges that significantly affect the reliability of AI feature extraction and comparison. The three ranges formed by the dual thresholds of the first and second grayscale values ​​cover all scenarios—stable and usable, critically unverifiable, and completely unusable—and allow for rapid decision-making through clear boundary values, meeting the real-time priority requirements of industrial-grade AI systems. The preset first and second grayscale values ​​can be set according to actual working conditions. The setting of the preset first and second grayscale values ​​aims to ensure the stability and practicality of AI image comparison analysis. Optionally, the preset first and second grayscale values ​​are determined through a limited number of experiments by evaluating the analytical effect of different grayscale levels on AI image comparison. The determined preset first and second grayscale values ​​should be neither too small nor excessively interfere with the AI ​​image comparison analysis process. For example, the preset range of the first grayscale value is generally [130, 170], and the preset range of the second grayscale value is generally [180, 220].

[0065] Preferably, the first grayscale value is preset to 150 in a preferred embodiment, and the second grayscale value is preset to 200 in a preferred embodiment.

[0066] Specifically, the average grayscale value of a craft image is the ratio of the total grayscale value of a number of craft images to the number of craft images.

[0067] In practice, the system described in this application determines the analytical stability of AI image comparison by setting a preset first grayscale value and a preset second grayscale value, thereby reducing the impact of the decrease in the analytical accuracy of AI image comparison due to the inaccuracy in determining the analytical stability of AI image comparison, and further improving the analytical stability of AI image comparison.

[0068] Specifically, the light intensity adjustment module increases the light intensity weighting coefficient of the underexposed area in response to the average gray value of the craft image being less than or equal to the preset first gray value.

[0069] The increase in the light intensity weighting coefficient of the underexposed area is determined by the difference between a preset first gray value and the average gray value of the craft image.

[0070] Specifically, when the difference between the preset first gray value and the average gray value of the craft image is within 20, the light intensity weighting coefficient of the underexposed area is increased to 1.1 times the original value. When the difference between the preset first gray value and the average gray value of the craft image exceeds 20, in addition to increasing to 1.1 times the original value, the light intensity weighting coefficient of the underexposed area increases by 0.5 for every 5 values ​​exceeding 20. For example, if the difference between the preset first gray value and the average gray value of the craft image is 30, and the current light intensity weighting coefficient of the underexposed area is 2, the increased light intensity weighting coefficient of the underexposed area is 2×1.1+0.5×2=3.2.

[0071] In practice, the system described in this application adjusts the light intensity weighting coefficient of the underexposed area by setting a preset first gray value and a preset second gray value. Due to the slight displacement of the camera caused by the vibration of the workshop equipment, the angle of light received by the lens changes, and the amount of light entering the camera decreases. The parameters are not adjusted synchronously, resulting in underexposure and increased noise. By increasing the light intensity weighting coefficient of the underexposed area, the brightness of the underexposed area can be increased first, accurately filling the light intensity gap caused by the reduction in the amount of light entering the camera. This avoids overexposure of normal areas in order to compensate for underexposure, and further improves the analysis stability of AI image comparison.

[0072] Specifically, the frequency adjustment module responds to the transmission delay of the craft image being less than or equal to a preset first delay duration, and determines that the signal integrity of the craft image meets the requirements.

[0073] The frequency adjustment module determines that the signal integrity of the craft image does not meet the requirements when the transmission delay of the craft image is greater than the preset first delay duration.

[0074] Specifically, the frequency adjustment module responds to the transmission delay of the craft image being greater than the preset first delay duration and less than or equal to the preset second delay duration by reducing the signal frequency jump interval of the craft image;

[0075] The frequency adjustment module responds to the fact that the transmission delay of the craft image is greater than the preset second delay duration, initially determines that the encoding accuracy of the craft image does not meet the requirements, and determines whether the encoding accuracy of the craft image meets the requirements based on the average encoding duration of several frames of craft images.

[0076] It is understandable that the preset first delay duration is shorter than the preset second delay duration, and the three intervals divided by the preset first delay duration and the preset second delay duration correspond to three different scenarios:

[0077] The first interval is when the transmission delay of the craft image is less than or equal to the preset first delay time, which corresponds to the situation where the signal integrity of the craft image meets the requirements.

[0078] The second interval is when the transmission delay of the craft image is greater than the preset first delay and less than or equal to the preset second delay. The corresponding situation is that, since multiple sets of equipment may share the same transmission network, signal superposition interference may occur, which will lead to a decrease in transmission rate.

[0079] The third interval is when the transmission delay of the craft image is longer than the preset second delay. The corresponding situation is: since the AI ​​image recognition system needs to process image acquisition, compression encoding, feature extraction, comparison and analysis, etc. at the same time, if multiple tasks compete for the same hardware resources, it will lead to insufficient resources of the compression encoding component.

[0080] Understandably, the preset first and second delay durations characterize the image quality of the artwork. The core logic is that there is an indirect but strong causal relationship between transmission delay and image quality. Dividing the image into acceptable, optimization-needed, and high-risk ranges using two delay thresholds aligns with the technical principles of image transmission and provides a layered guarantee for the stability of AI comparison. The preset first delay duration defines the critical line for acceptable quality, filtering out high-quality images that can be directly used for AI comparison. The preset second delay duration defines the escalation line for quality risk, distinguishing between minor, repairable risks and severe risks requiring in-depth verification. The preset first and second delay durations can be set according to actual working conditions. The setting of the preset first and second delay durations aims to ensure the stability and practicality of AI image comparison analysis. Optionally, the preset first and second delay durations are determined through a limited number of experiments by evaluating the analytical effects of different delay durations on AI image comparison. The determined preset first and second delay durations should be neither too small nor excessively interfere with the AI ​​image comparison analysis process. For example, the preset first delay duration is generally selected in the range of [400ms, 600ms], and the preset second delay duration is generally selected in the range of [700ms, 900ms].

[0081] Preferably, the first delay duration is 500ms in a preferred embodiment, and the second delay duration is 800ms in a preferred embodiment.

[0082] In practice, the system described in this application determines the signal integrity of the craft image by setting a preset first delay duration and a preset second delay duration, thereby reducing the impact of inaccurate determination of the signal integrity of the craft image on the analysis stability of AI image comparison and further improving the analysis stability of AI image comparison.

[0083] Specifically, the reduction in the signal frequency switching interval of the craft image is determined by the difference between the transmission delay of the craft image and a preset first delay.

[0084] Specifically, when the difference between the transmission delay of the craft image and the preset first delay is within 50ms, the signal frequency switching interval of the craft image is reduced to 0.95 times its original value. When the difference between the transmission delay of the craft image and the preset first delay exceeds 50ms, in addition to reducing it to 0.95 times its original value, the signal frequency switching interval of the craft image is reduced by 50ms for every 20ms exceeding the original value. For example, if the difference between the transmission delay of the craft image and the preset first delay is 90ms, and the current signal frequency switching interval of the craft image is 700ms, the reduced signal frequency switching interval of the craft image is 700×0.95-50×2=565ms.

[0085] In practice, the system described in this application adjusts the signal frequency switching interval of the craft image by setting a preset first delay duration and a preset second delay duration. Since multiple devices may share the same transmission network, signal superposition interference may occur, resulting in a decrease in transmission rate. By reducing the signal frequency switching interval of the image, the device can stay on a single frequency for a shorter time, reducing the probability of being hit by interference signals. This allows the system to quickly escape occupied channels, seize clean frequencies to transmit image data, reduce transmission interruptions caused by interference, indirectly improve the effective transmission rate, and further enhance the analytical stability of AI image comparison.

[0086] Specifically, the quantity adjustment module determines that the encoding accuracy of the handicraft image meets the requirements when the average encoding time of several frames of handicraft images is less than or equal to the preset encoding time.

[0087] The quantity adjustment module responds to the fact that the average encoding duration of the several frames of craft images is greater than the preset encoding duration, determines that the encoding accuracy of the craft images does not meet the requirements, and reduces the quantity threshold of intra-frame prediction modes.

[0088] It is understandable that the two intervals corresponding to the preset encoding duration represent two different scenarios:

[0089] The first interval is when the average encoding time of several frames of handicraft images is less than or equal to the preset encoding time, which corresponds to the situation where the encoding accuracy of the handicraft images meets the requirements.

[0090] The second interval is when the average encoding time of several frames of craft images is greater than the preset encoding time. The corresponding situation is: since the AI ​​image recognition system needs to process image acquisition, compression encoding, feature extraction, comparison and analysis, etc. at the same time, if multiple tasks compete for the same hardware resources, it will lead to insufficient resources of the compression encoding component.

[0091] Understandably, in AI image recognition-based comparison and analysis systems, encoding duration is used to characterize the encoding accuracy of handicraft images. The core logic is that encoding duration is strongly correlated with the completeness and standardization of the encoding process. Under normal circumstances, encoding that meets completeness requirements will follow a preset process (such as intra-frame prediction, transform quantization, entropy encoding, and other standard steps), and its time consumption will fall within a reasonable range. If the encoding duration abnormally exceeds the threshold, it often means that the encoding process has not been completely executed, ultimately resulting in incomplete image data, which in turn affects the accuracy of subsequent AI recognition and comparison. The preset encoding duration can be set according to actual working conditions. The setting of the preset encoding duration aims to ensure the stability and practicality of AI image comparison and analysis. Optionally, the preset encoding duration is determined through a limited number of experiments by evaluating the analytical effect of different encoding durations on AI image comparison. The determined preset encoding duration should be neither too small nor cause excessive interference to the AI ​​image comparison and analysis process. For example, the preset encoding duration is generally selected within the range of [15ms, 25ms].

[0092] Preferably, the preferred embodiment of the preset encoding duration is 20ms.

[0093] Specifically, the average encoding time of a number of craft images is the ratio of the total encoding time of the number of craft images to the number of craft images.

[0094] In practice, the system described in this application determines the encoding accuracy of the handicraft image by setting a preset encoding duration, thereby reducing the impact of inaccurate determination of the encoding accuracy of the handicraft image on the analysis stability of AI image comparison, and further improving the analysis stability of AI image comparison.

[0095] Specifically, the reduction in the number threshold of the intra-frame prediction mode is determined by the difference between the average encoding time of several frames of craft images and the preset encoding time.

[0096] Specifically, when the difference between the average encoding time of several frames of craft images and the preset encoding time is within 5ms, the threshold for the number of intra-frame prediction modes is reduced to 0.9 times the original value. When the difference between the average encoding time of several frames of craft images and the preset encoding time exceeds 5ms, in addition to reducing it to 0.9 times the original value, the threshold for the number of intra-frame prediction modes is reduced by 1 for every 1ms exceeding the original value. For example, if the difference between the average encoding time of several frames of craft images and the preset encoding time is 7ms, and the current threshold for the number of intra-frame prediction modes is 30, the reduced threshold for the number of intra-frame prediction modes is 30 × 0.9 - 1 × 2 = 25.

[0097] In practice, the system described in this application adjusts the threshold for the number of intra-frame prediction modes by setting a preset encoding duration. Since the AI ​​image recognition system needs to process tasks such as image acquisition, compression encoding, feature extraction, and comparison analysis simultaneously, if multiple tasks compete for the same hardware resources, it will lead to insufficient resources for the compression encoding component. By reducing the threshold for the number of intra-frame prediction modes, the amount of computation can be reduced, the demand for resources can be lowered, and the compression encoding component can work more efficiently with limited resources, reducing processing time and further improving the stability of AI image comparison analysis.

[0098] A comparative analysis method based on AI image recognition includes:

[0099] Step S1: Use a camera to capture images of handicrafts and transmit them to the processing location. Perform noise reduction, cropping, encoding, and feature extraction operations on the handicraft images in sequence to output feature images. Train the initial model based on the feature images to output a machine learning model.

[0100] Step S2: Use the machine learning model to compare the handicraft image with the reference image to output image similarity, and analyze the image similarity to output analysis results;

[0101] Step S3: Obtain the grayscale values ​​of several craft images, calculate the average grayscale value of the craft images, and determine whether the analysis stability of the AI ​​image comparison meets the requirements based on the average grayscale value of the craft images.

[0102] Step S4: If the analysis stability of the AI ​​image comparison does not meet the requirements, determine whether it is necessary to increase the light intensity weighting coefficient of the underexposed area.

[0103] Step S5: If it is not necessary to increase the light intensity weighting coefficient of the underexposed area, then obtain the transmission delay time of the craft image to determine whether the signal integrity of the craft image meets the requirements.

[0104] Step S6: If the signal integrity of the craft image does not meet the requirements, determine whether it is necessary to reduce the signal frequency transition interval of the craft image.

[0105] Step S7: If it is not necessary to reduce the signal frequency jump interval of the craft image, then determine the threshold of the number of intra-frame prediction modes based on the average coding time of several frames of craft images.

[0106] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.

Claims

1. A comparison and analysis system based on AI image recognition, characterized in that, include: The image acquisition module includes an image acquisition unit for acquiring images of handicrafts via a camera and an image transmission unit connected to the image acquisition unit for transmitting the images of handicrafts to a processing location; An image processing module, connected to the image acquisition module, includes a preprocessing unit for sequentially performing denoising, cropping, encoding, and feature extraction operations on the handicraft image to output a feature image, and a model training unit connected to the preprocessing unit for training an initial model based on the feature image to output a machine learning model. The comparison and analysis module, which is connected to the image processing module, includes a comparison unit for comparing the handicraft image with a reference image using the machine learning model to output image similarity, and an analysis unit connected to the comparison unit for analyzing the image similarity to output analysis results. A light intensity adjustment module, which is connected to the image acquisition module, is used to determine the light intensity weighting coefficient of the underexposed area based on the average gray value of the handicraft image. A frequency adjustment module, which is connected to the image acquisition module and the light intensity adjustment module respectively, is used to determine the signal frequency jump interval of the handicraft image based on the transmission delay time of the handicraft image; The quantity adjustment module, which is connected to the image processing module and the frequency adjustment module respectively, is used to determine the quantity threshold of the intra-frame prediction mode based on the average encoding time of several frames of craft images.

2. The comparison and analysis system based on AI image recognition according to claim 1, characterized in that, The light intensity adjustment module responds to the fact that the average gray value of the craft image is greater than or equal to a preset second gray value, thus determining that the analysis stability of the AI ​​image comparison meets the requirements. The light intensity adjustment module determines that the analysis stability of the AI ​​image comparison does not meet the requirements when the average gray value of the craft image is less than the preset second gray value.

3. The comparison and analysis system based on AI image recognition according to claim 2, characterized in that, The light intensity adjustment module responds to the fact that the average gray value of the craft image is greater than a preset first gray value and less than a preset second gray value, and preliminarily determines that the signal integrity of the craft image does not meet the requirements.

4. The comparison and analysis system based on AI image recognition according to claim 3, characterized in that, The light intensity adjustment module increases the light intensity weighting coefficient of the underexposed area in response to the average gray value of the craft image being less than or equal to the preset first gray value. The increase in the light intensity weighting coefficient of the underexposed area is determined by the difference between a preset first gray value and the average gray value of the craft image.

5. The comparison and analysis system based on AI image recognition according to claim 3, characterized in that, The frequency adjustment module responds to the transmission delay of the craft image being less than or equal to a preset first delay duration, and determines that the signal integrity of the craft image meets the requirements. The frequency adjustment module determines that the signal integrity of the craft image does not meet the requirements when the transmission delay of the craft image is greater than the preset first delay duration.

6. The comparison and analysis system based on AI image recognition according to claim 5, characterized in that, The frequency adjustment module responds to the transmission delay of the craft image being greater than the preset first delay and less than or equal to the preset second delay by reducing the signal frequency jump interval of the craft image. The frequency adjustment module responds to the fact that the transmission delay of the craft image is greater than the preset second delay duration, and preliminarily determines that the encoding accuracy of the craft image does not meet the requirements.

7. The comparison and analysis system based on AI image recognition according to claim 6, characterized in that, The reduction in the signal frequency transition interval of the craft image is determined by the difference between the transmission delay of the craft image and a preset first delay.

8. The comparison and analysis system based on AI image recognition according to claim 7, characterized in that, The quantity adjustment module responds to the fact that the average encoding time of several frames of craft images is less than or equal to the preset encoding time, and determines that the encoding accuracy of the craft images meets the requirements. The quantity adjustment module responds to the fact that the average encoding duration of the several frames of craft images is greater than the preset encoding duration, determines that the encoding accuracy of the craft images does not meet the requirements, and reduces the quantity threshold of intra-frame prediction modes.

9. The comparison and analysis system based on AI image recognition according to claim 8, characterized in that, The reduction in the number threshold of the intra-frame prediction mode is determined by the difference between the average encoding time of several frames of craft images and the preset encoding time.

10. A comparison analysis method applied to the comparison analysis system based on AI image recognition as described in any one of claims 1-9, characterized in that, include: The image of the handicraft is captured by a camera and transmitted to the processing location. The handicraft image is then subjected to noise reduction, cropping, encoding and feature extraction operations to output a feature image. The initial model is trained based on the feature image to output a machine learning model. The machine learning model is used to compare the image of the handicraft with a reference image to output image similarity, and the image similarity is analyzed to output analysis results; The grayscale values ​​of several handicraft images are obtained, and the average grayscale value of the handicraft images is calculated. Based on the average grayscale value of the handicraft images, it is determined whether the analysis stability of the AI ​​image comparison meets the requirements. If the analysis stability of the AI ​​image comparison does not meet the requirements, it is determined whether the light intensity weighting coefficient of the underexposed area needs to be increased. If it is not necessary to increase the light intensity weighting coefficient of the underexposed area, then obtain the transmission delay time of the craft image to determine whether the signal integrity of the craft image meets the requirements. If the signal integrity of the craft image does not meet the requirements, determine whether it is necessary to reduce the signal frequency transition interval of the craft image. If it is not necessary to reduce the signal frequency jump interval of the craft images, the number threshold of intra-frame prediction modes is determined based on the average coding duration of several frames of craft images.

Images