Research On Key Technologies For Classification And Identification Of Common Pediatric Skin Disease

The quality and efficiency of medical technical services are tangibly related to the health and well-being of every people.As a common disease,the traditional diagnosis of pediatric dermatoses is mainly based on doctors’ clinical experience,and its effect is easily limited by the shortage and experience level of clinical specialists in pediatric dermatology,making it impossible for patients to obtain a high-quality medical experience.With the continuous penetration and integration of artificial intelligence technology in the medical field,using appropriate classification and recognition technology can effectively assist physicians make objective and repeatable clinical diagnoses and decisions,improve diagnostic accuracy and efficiency and enhance the quality of primary care services.Combining the objective needs of characterization and detection of common pediatric skin diseases and the characteristics of pediatric skin diseases in natural scenes.This article focuses on the problems of the challenge of image collection,unstable image quality,the weak performance of small objects,the intra-class variability and inter-class similarity of disease symptoms,and the mixing of disease symptom detection results in natural scenes.CF-DKFD algorithm based on the Faster R-CNN is proposed,which improves the performance of disease representation detection,and the main content is as follows.(1)For the issue of the small dataset and low resolution of images,the Random Online Data Augmentation and Selective Image Super-Resolution Reconstruction(RDA-SSR)method is introduced,which successfully alleviate the overfitting in training,and improves the performance of disease detection.(2)For the poor performance of small object detection,the CF-PVTv2 feature fusion backbone network is proposed based on the PVTv2.Which fuses deep semantic information with shallow detail features,and uses channel attention to establish the relationship between high-level semantic and spatial information channels.(3)For the issue of an imbalance between difficult and simple samples brought on by the variation within and between classes of disease signs during distinct disease phases.A new loss function for two-stage object detection named DK＿Loss is proposed,by increasing the loss contribution of hard samples,allowing the model to concentrate more on learning hard samples within the normal range.(4)For the redundancy detection results of the disease representation,a new postprocessing method Fliter＿nms is proposed,to reduce significantly incorrect detection boxes.We created the CPD-10 image dataset for common pediatric dermatoses and used the Faster R-CNN network training findings as a benchmark,and experimentally verified the effectiveness of the CF-DKFD algorithm and the generalizability of CFPVTv2 and DK＿Loss.The experimental results confirmed that over the Faster R-CNN,the CF-DKFD algorithm proposed in this paper can improve the m AP of pediatric common dermatology symptom detection by over 7%,and effectively improving the precision of the model.