Zero-sample learning method based on data enhancement

Abstract

The invention discloses a zero-sample learning method based on data enhancement. The zero-sample learning method comprises the steps of drawing a contour of a focus, performing focus contour processing on the contour according to a morphological operation method, and then filling a color on the focus by means of a continuous gradient color filling method; increasing diversity of the focus with filled colors by means of a matrix linear transforming method, generating a plurality of simulated focus pictures, and combining the simulated focus pictures with a disease background picture by means ofimage fusion technology, and adding a Gaussian white noise to the fused picture for obtaining an expanded sample set; training a VGG classifier by means of the expanded sample set, and training an optimal classifier; and testing a dermatopathy case by means of the optimal classifier. The zero-sample learning method effectively settles a zero-sample learning auxiliary diagnosis problem of a rare disease.

Description

technical field [0001] The invention relates to the field of medical image aided diagnosis, in particular to a zero-sample learning method based on data enhancement. Background technique [0002] At present, medical resources are scarce, professional doctors are in short supply, and the doctor training cycle is long; medical resources are distributed unevenly, large hospitals are overcrowded, and small hospitals are neglected. The application of deep learning in medicine can help medically underdeveloped regions and hospitals to develop intelligent medical diagnostic robots; provide professional and accurate diagnostic assistance, improve the level of medical diagnosis, and reduce the rate of misdiagnosis; reduce the work pressure of doctors and improve the efficiency of medical workers. work efficiency. However, the current medical image-aided diagnosis has the following problems: most of the medical image-aided diagnosis at the current stage is supervised learning, and a ...

AI Technical Summary

Problems solved by technology

However, the current medical image-aided diagnosis has the following problems: most of the medical image-aided diagnosis at the current stage is supervised learning, and a large amount of labeled medical image data needs to be collected to train the model

In this actual scenario, the novice doctor does not see the real picture of the disease, and can identify the disease only based on the description of the disease outline and other attributes by the expert doctor. The current zero-shot learning does not meet the novice doctor's understanding of the disease. learning cognitive process

Images