Search

CN-122022794-A - Picosecond laser speckle removing charging method and system based on neural network

CN122022794ACN 122022794 ACN122022794 ACN 122022794ACN-122022794-A

Abstract

The invention relates to the field of image processing, in particular to a picosecond laser speckle removing charging method and system based on a neural network. The method comprises the steps of obtaining a facial image of a patient to be treated, carrying out image enhancement and background separation processing on the facial image to obtain an image to be detected which only retains skin characteristics, inputting the image to be detected into a pre-trained semantic segmentation neural network model to obtain a pixel-level mask image for identifying a color spot area, counting the total number of pixel points identified as the color spot area, calculating the charging number of the color spot area converted into a standard charging unit, obtaining the price corresponding to a single standard charging unit corresponding to the current color spot type, and generating final treatment cost based on the charging number and the price corresponding to the single standard charging unit. The method can realize objectification and automation of charging for the treatment cost of facial color spots in picoseconds.

Inventors

  • LIU LIQIANG

Assignees

  • 广州市中崎商业机器股份有限公司

Dates

Publication Date
20260512
Application Date
20260204

Claims (10)

  1. 1. The picosecond laser speckle removing charging method based on the neural network is characterized by comprising the following steps of: acquiring a facial image of a patient to be treated, wherein the facial image is acquired under the preset imaging distance and illumination conditions; Performing image enhancement and background separation processing on the facial image to obtain an image to be detected, wherein only skin characteristics are reserved; inputting the image to be detected into a pre-trained semantic segmentation neural network model, so as to obtain a pixel-level mask image for marking a color spot area; traversing the pixel-level mask image, counting a total number of pixels identified as a stain region; Calculating the total number of pixel points in the color spot area and the number of pixel points corresponding to a single standard charging unit, and converting the color spot area into the charging number of the standard charging unit; and acquiring the price corresponding to the single standard charging unit corresponding to the current color spot type, and generating the final treatment expense based on the charging quantity and the price corresponding to the single standard charging unit.
  2. 2. The neural network-based picosecond laser speckle-removing billing method of claim 1, wherein the image enhancement and background separation of the facial image comprises: performing illumination component estimation and removal on the face image to obtain an illumination normalized image; carrying out noise reduction treatment on the illumination normalization image to obtain a noise reduction image; Converting the noise reduction image into a YCbCr color space, and extracting Cr channel data; and judging whether the pixel value in the Cr channel data is in a preset skin color threshold interval, reserving pixels in the skin color threshold interval, removing background pixels outside the interval, and generating the image to be detected.
  3. 3. The method for picosecond laser speckle removal and charging based on neural network as claimed in claim 2, wherein the face image is estimated and removed by using Retinex algorithm, and the illumination normalized image is noise-reduced by using bilateral filtering algorithm with kernel size of preset value.
  4. 4. The neural network-based picosecond laser speckle-removing billing method of claim 1 wherein calculating the billing amount for the conversion of the speckle region to a standard billing unit comprises: calculating the ratio R of the color spot area to the number of pixel points of a standard charging unit, wherein the calculation expression is as follows: ; Wherein N represents the total number of pixel points of the color spot area, and S represents the number of pixel points of the standard charging unit; extracting an integer part I and a decimal part D of the area ratio R; Judging whether the decimal part D is greater than or equal to a preset judging threshold value ; If the decimal part D is greater than or equal to Determining that the charging quantity Q is equal to I+1, if the decimal part D is smaller than And judging that the charging quantity Q is equal to I.
  5. 5. The neural network-based picosecond laser speckle-removing billing method of claim 4 wherein the predetermined decision threshold value The value is 0.5.
  6. 6. The picosecond laser speckle removing and charging method based on the neural network as set forth in claim 1, wherein the number of pixels corresponding to the single standard charging unit is calculated by: Acquiring a camera imaging calibration coefficient K and a physical side length L of a standard charging unit, wherein K represents the number of pixels corresponding to the unit physical length; Calculating the number of pixel points corresponding to a single standard charging unit according to the calibration coefficient K and the physical side length L The computational expression is: 。
  7. 7. the neural network-based picosecond laser speckle-removing billing method of claim 1 further comprising a camera calibration step prior to obtaining the camera imaging calibration coefficient K, comprising: Acquiring a plurality of calibration images containing standard physical size calibration blocks; processing the calibration image by using a camera calibration algorithm, and calculating a focal length parameter of a camera; and calculating the number of pixels mapped on the image sensor in a unit physical length based on the focal length parameter and the preset imaging distance, determining the number of pixels as the calibration coefficient K and storing the calibration coefficient K.
  8. 8. The picosecond laser speckle removing charging method based on the neural network according to any one of claims 1-7, wherein the semantic segmentation neural network model is optimized through migration learning, and the training process comprises the following steps: Constructing a labeling image data set containing different skin color types and different color spot types, wherein pixel-level labeling is carried out on color spot boundaries in the labeling image; And carrying out iterative training on the basic convolutional neural network by using the labeling image data set until the average Dice coefficient of the model for identifying the boundaries of the color spots reaches a preset precision threshold, keeping the trained model parameters and obtaining the semantic segmentation neural network model, wherein the boundary error is not larger than the preset error.
  9. 9. The utility model provides a picosecond laser speckle removing charging system based on neural network which characterized in that includes: the high-definition imaging module is used for acquiring facial images under a fixed imaging distance and constant illumination environment; The computing power processing module is connected with the high-definition imaging module and is used for executing the picosecond laser speckle removing charging method based on the neural network according to any one of claims 1 to 8; and the interaction output module is connected with the computing power processing module and is used for displaying the segmentation result of the pixel-level mask image, displaying the charging quantity Q and the final treatment cost and printing a certificate containing the certificate storage information.
  10. 10. The neural network-based picosecond laser speckle-removing billing system of claim 9 wherein the high definition imaging module comprises a CMOS camera configured with an annular uniform illumination assembly and a face positioning reference for limiting the distance of the patient's face to be treated from the CMOS camera lens to within a preset range.

Description

Picosecond laser speckle removing charging method and system based on neural network Technical Field The present invention relates to the field of image processing. More particularly, the invention relates to a picosecond laser speckle removing charging method and system based on a neural network. Background The picosecond laser technology has become a mainstream means for clinically treating facial pigment diseases such as freckle, chloasma, senile plaque and the like because of short pulse width and high peak power and capability of accurately vibrating and crushing skin melanin. In the actual diagnosis and treatment process of medical cosmetology, the treatment cost calculation aiming at scattered and irregular color spots is a key link. Unlike uniform pricing of full face patterns, treatment for localized spots generally requires a fee to be calculated from the actual coverage area of the spot. However, the prior art relies primarily on visual inspection and manual evaluation by the physician when performing area metering and charging for such facial spots. In practice, doctors generally use a subjective reference (such as "nail cover size", "coin size") as a price unit to roughly accumulate scattered color spots with different shapes on the face of the patient. This non-standardized mode of operation, which relies on manual experience, has significant technical drawbacks: First, accuracy and objectivity of area quantization are insufficient. The stain typically exhibits a characteristic of blurring edges, extremely irregular shapes (e.g., cloudiness, map-like), and often in a scattered distribution. The human eye has difficulty in accurately measuring and calculating the physical area of the tiny areas with non-geometric shapes, so that the estimation result often has larger deviation. This quantification based on subjective visual perception lacks uniform physical metrics, resulting in the possibility of significant differences in the area values obtained by the same patient in different doctors and even in different evaluations of the same doctor. Second, the accounting efficiency of the complex-form stain is low. For patients with a large number of fine freckles or large area diffuse chloasma on the face, a large amount of diagnosis and treatment time is required to be consumed by manually identifying and accumulating the estimated areas one by one. In a busy outpatient environment, the inefficient metering mode not only occupies medical operation time of doctors, but also prolongs waiting period of patients, and influences overall diagnosis and treatment service flux. In addition, although the existing skin image AI auxiliary technology is applied to classification diagnosis of skin diseases (such as judging color spot properties and screening malignant melanoma), the technical key point is qualitative classification of image features, and a complete technical scheme for performing precise physical area measurement on skin damage areas and directly converting the precise physical area measurement into charging data is not formed. The existing general image processing software has a pixel statistics function, but lacks a special pretreatment mechanism aiming at the complicated illumination environment of the face and the interference of the complexion background, does not establish standardized mapping logic between pixel coordinates and a physical charging unit, and cannot be directly applied to medical charging scenes. In summary, the existing charging method for the treatment of facial spots and picoseconds has the technical problems of lack of objective and unified physical quantization standard, low measurement accuracy for irregular micro-areas and low manual accounting efficiency, and a technical scheme capable of automatically and accurately carrying out physical quantization on the areas of the spots and completing standardized charging is needed. Disclosure of Invention In order to solve the technical problems that the existing charging mode for the facial stain picosecond treatment lacks objective and unified physical quantization standard, has low measurement precision on irregular micro-areas and has low manual accounting efficiency, the invention provides a scheme in the following aspects. In a first aspect, the present invention provides a picosecond laser speckle removing charging method based on a neural network, including: acquiring a facial image of a patient to be treated, wherein the facial image is acquired under the preset imaging distance and illumination conditions; Performing image enhancement and background separation processing on the facial image to obtain an image to be detected, wherein only skin characteristics are reserved; inputting the image to be detected into a pre-trained semantic segmentation neural network model, so as to obtain a pixel-level mask image for marking a color spot area; traversing the pixel-level mask image, counting a total number of pixels iden