Search

CN-122025073-A - Image optimization auxiliary method based on AI (advanced technology attachment) recognition and diagnosis analysis system

CN122025073ACN 122025073 ACN122025073 ACN 122025073ACN-122025073-A

Abstract

The invention relates to the field of medical image optimization, in particular to an image optimization auxiliary method and a diagnosis analysis system based on AI identification, which are used for acquiring an original diagnosis image obtained through shooting, carrying out noise reduction treatment on the original diagnosis image, dividing the contour of a cut surface or tongue of the image subjected to the noise reduction treatment into a plurality of areas, dividing the contour of the face or tongue into a plurality of areas, preprocessing pixels at key positions, analyzing the preprocessed areas and carrying out edge analysis on different positions to obtain an analyzable gray level image, training the diagnosis image through a convolutional neural network through the analyzable gray level image, and outputting a diagnosis image processing result.

Inventors

  • LING LIN

Assignees

  • 广州德生智能信息技术有限公司

Dates

Publication Date
20260512
Application Date
20251218

Claims (6)

  1. 1. An image optimization assisting method based on AI recognition, which is characterized by comprising the following steps: S100, acquiring an original diagnostic image obtained by shooting, and carrying out noise reduction treatment on the original diagnostic image; s200, cutting the section or tongue outline of the image subjected to noise reduction treatment, and dividing the acquired section or tongue outline into a plurality of areas; s300, preprocessing key position pixel points of a plurality of areas divided by the facial or tongue outline; s400, analyzing the preprocessed region and carrying out marginalization analysis on different positions to obtain an analyzable gray level image; And S500, training the diagnostic image through a convolutional neural network through the analyzable gray image, and outputting a diagnostic image processing result.
  2. 2. The AI-recognition-based image optimization assisting method according to claim 1, wherein in step S100, an original diagnostic image of a user is acquired through a camera in a diagnostic system, the original diagnostic image includes a face image and a tongue image, the face image and the tongue image are subjected to brightness processing, the acquired original diagnostic image is subjected to light supplementing according to the current environment of the user, color restoration of the original diagnostic image is ensured, and noise reduction processing is performed on the original diagnostic image, so that face or tongue details of the original diagnostic image can be ensured to be acquired.
  3. 3. The AI recognition-based image optimization auxiliary method as set forth in claim 1, wherein in step S200, the original diagnostic image is divided into a plurality of subareas on average, and key positions in each subarea are collected to form a digital twin data set; And displaying the abnormal digital twin area at a corresponding position in the three-dimensional model of the preset area to form a visual three-dimensional image display.
  4. 4. The AI-recognition-based image optimization assistance method as set forth in claim 3, wherein in step S300, the time when the difference pixel occurs is recorded, the position density of the difference pixel occurs is divided, the pixel density is calculated for a plurality of position division areas with high pixel density, the areas of the division areas are the same, the pixel density DP is obtained by the total number of the difference pixel occurring in the division areas/the area of the division areas, the average value att of euclidean distances between the software end and the hardware end of the pixel where the difference pixel occurs in all the division areas is calculated, and the distortion deviation coefficient in the division areas is calculated by the pixel density and the average value att of euclidean distances between the software end and the hardware end of the pixel where the difference pixel occurs: Pf(l(Di))= * ; Wherein Di is an independent variable in the formula, di is denoted as an occurrence difference pixel point with a number i, l (Di) is denoted as Euclidean distance between a software end and a hardware end of the pixel point with the occurrence difference pixel point with the number i, DPe is denoted as pixel point density DP of heavy dividing regions with a number e, the pixel point Di is in the dividing regions with a number e, pf (l (Di)) is a function for calculating distortion deviation coefficients, Expressed as a ratio of the degree of distortion to the concentration under the divided region, The minimum value of Euclidean distance between a software end and a hardware end of the pixel points with difference appears, and the distortion deviation coefficient of the pixel points Di in the dividing region e is obtained by carrying out calculation on an independent variable l (Di) in the formula; S303, determining a distortion deviation direction, determining a distortion direction from a software end to a hardware end according to a pixel point software end and a hardware end of a pixel point with difference, and marking as Recording the distortion directions from the software end to the hardware end of all the pixels with difference in the dividing area e, and obtaining the vector sum to obtain the overall distortion direction of the dividing area e ; The transmission efficiency of information transmission between a software end and a hardware end is reflected by the number S of image outputs, wherein ENi=S/ti, ti is the time of the number S of projection images output, i is the recorded ith time period, EN= [ ENi ], the total number of elements is m, EN is the set of the transmission efficiency ENi of information transmission between the software end and the hardware end which are ordered according to time sequence, and the information frequency correction value XEN, XENi=is obtained through calculation ) T is the total duration, the information frequency correction value XEN is the ratio of the transmission efficiency ENi at the moment i to the overall transmission efficiency, and the information frequency correction value XEN can be combined with the time period for generating distortion to judge, so that the influence between the distortion and the transmission speed is generated; acquiring an arithmetic mean avg (XENi) of the information frequency correction value, wherein avg () is a function for acquiring the arithmetic mean in the array; Defining XENi > avg (XENi) as one occurrence of abnormal information, judging that the transmission time of the abnormal information in the projection of the software end to the hardware end is j, recording the number of times of recording the abnormal information in the projection process of the software end to the hardware end, recording k as the number of times of recording the abnormal information, combining and judging the distortion deviation coefficient and the distortion deviation direction of the pixel point with the difference in the occurrence of the abnormal information, obtaining the distortion deviation coefficient and the distortion deviation direction of each divided area in the j time, T1= + * j( ); T2= * ; g=( ); Where T1 is represented as a distortion deviation of the divided region at the time j, T2 is represented as a distortion deviation of the pixel point Di at the time j, g includes a sum of the information frequency correction values XEN, T1, T2 and g are passed through a function E (), E()=μg , Wherein μ is a distortion error, and is a constant value, exp () is an exponential function, and if the value of the function E () is equal to or greater than 1, it indicates that there is distortion due to transmission efficiency at time j, and if the value of the function E () is <1, it indicates that no distortion deviation occurs due to transmission rate at time j.
  5. 5. The AI-recognition-based image optimization supporting method according to claim 1, wherein in step S400, the image diagnosis model is generated by learning teaching data that is a plurality of medical images different from the device data, and the adjustment unit generates a teaching histogram that is a histogram of the size of a tumor contained in each medical image of the teaching data, and adjusts the size of the diagnostic image based on a device-most value that is a most-frequent value of the device histogram and a teaching-most value that is a most-frequent value of the teaching histogram; the adjustment unit multiplies the value obtained by dividing the teaching maximum frequency value by the device maximum frequency value by the size of the diagnostic image; the adjustment unit adjusts the diagnostic image by removing a high-luminance region in the vicinity of a tumor included in each medical image of the device data; a determination unit configured to determine whether or not adjustment by the adjustment unit is necessary based on a diagnosis result of the image diagnosis model, the adjustment unit adjusting a diagnosis image to be input to the image diagnosis model or the image diagnosis model when the determination unit determines that the adjustment is necessary; A screen for selecting an item for statistical processing of the device data is displayed.
  6. 6. An AI-recognition-based image optimization auxiliary diagnosis analysis system is characterized by comprising a model reading unit for reading an image diagnosis model that outputs a diagnosis result for an inputted medical image, i.e., a diagnosis image; And an adjustment section that adjusts a diagnostic image to be input to the image diagnostic model or the image diagnostic model based on the device data, the adjustment section generating a device histogram that is a histogram of a size of a tumor contained in each medical image of the device data, the image diagnostic model being adjusted by relearning using relearning data that is data generated based on the device histogram; The image diagnosis model is generated by learning teaching data that is a plurality of medical images different from the device data, and the adjustment unit extracts the relearning data from the teaching data based on the device histogram; the adjustment section calculates a probability function of extracting a medical image from the teaching data based on the device histogram, and extracts the relearning data using the probability function.

Description

Image optimization auxiliary method based on AI (advanced technology attachment) recognition and diagnosis analysis system Technical Field The invention relates to the field of medical image optimization, in particular to an image optimization auxiliary method based on AI identification and a diagnosis analysis system. Background With the increase of the life rhythm, unhealthy life forms gradually become the normalcy of partial population, such as staying up, binge eating, lack of physical exercise, etc. Further, the human body is uncomfortable to different degrees, and the accumulation of the conditions can reduce the physique of the human body and seriously damage the health of people. In addition, a series of diseases are caused by factors such as working environment, labor intensity and pressure intensity, so that in order to reduce the occurrence of such situations, disease prevention is particularly necessary, and self-diagnosis of health status is an excellent prevention mode. The current self-diagnosis modes of the health state mainly comprise color observation, pulse cutting, diagnosis according to physical feeling and reaction, and the common physical reaction such as alopecia, constipation, insomnia and dreaminess, and the like. The pulse feeling method needs to know and understand the pulse condition principle of traditional Chinese medicine deeply to finish diagnosis, is difficult for common individuals to realize, has hysteresis effect according to the self-diagnosis method of physical feeling and reaction, has poor diagnosis and prevention effect, is an effective and quick self-diagnosis method by observing facial color, is a color diagnosis method in traditional Chinese medicine inspection, and needs the related traditional Chinese medicine basic theory as an auxiliary diagnosis tool. Disclosure of Invention In view of the above limitations of the prior art, the present invention is directed to an image optimization supporting method and a diagnostic analysis system based on AI identification, so as to solve one or more technical problems in the prior art, and at least provide a beneficial choice or creation condition. An image optimization assisting method based on AI identification, the method comprising the steps of: S100, acquiring an original diagnostic image obtained by shooting, and carrying out noise reduction treatment on the original diagnostic image; s200, cutting the section or tongue outline of the image subjected to noise reduction treatment, and dividing the acquired section or tongue outline into a plurality of areas; s300, preprocessing key position pixel points of a plurality of areas divided by the facial or tongue outline; s400, analyzing the preprocessed region and carrying out marginalization analysis on different positions to obtain an analyzable gray level image; And S500, training the diagnostic image through a convolutional neural network through the analyzable gray image, and outputting a diagnostic image processing result. Further, in step S100, an original diagnostic image of the user is collected by a camera in the diagnostic system, where the original diagnostic image includes a face image and a tongue image, the face image and the tongue image are subjected to brightness processing, and the collected original diagnostic image is supplemented according to the current environment of the user to ensure color restoration of the original diagnostic image, and then noise reduction processing is performed on the original diagnostic image to ensure that the face or tongue details of the original diagnostic image can be obtained. Further, in step S200, the original diagnostic image is divided into a plurality of sub-regions on average, and key positions in each sub-region are collected to form a digital twin dataset; calculating overflow values between each sub-region and each corresponding digital twin region, and marking out an abnormal constant region in the digital twin region of each sub-region through the overflow values; And displaying the abnormal digital twin area at a corresponding position in the three-dimensional model of the preset area to form a visual three-dimensional image display. Further, in step S300, the time when the difference pixel points appear is recorded, the position density of the difference pixel points appears is divided, the pixel point density is calculated for a plurality of position division areas with high pixel density, the areas of the division areas are the same, the pixel point density DP is obtained by the total number of the difference pixel points appearing in the division areas/the area of the division areas, the average value att of euclidean distances between the software end and the hardware end of the pixel points with the difference pixel points appearing in all the division areas is calculated, and the distortion deviation coefficient in the division areas is calculated by the pixel point density and the average value a