A Multi-Task Detection Method for Pixel-Level Segmentation of Surface Anomaly Regions

Abstract

The invention discloses a multi-task detection method for pixel-level segmentation of surface abnormal areas, which includes the following steps: acquiring an image data set, and performing pixel-level labeling on known abnormalities in it, and if not enough abnormal samples can be obtained, random Add noise as an anomaly; build a multi-task anomaly detection network model, which includes known anomaly detection branch D and unknown anomaly detection branch S; according to the pre-built multi-task optimization goal, train the multi-task anomaly detection network model through the image dataset , to obtain the final anomaly detection network model, in which the multi-task optimization target includes a pixel-level classification loss function and a compactness loss function; the image to be detected is input into the anomaly detection network model, and the known anomaly detection branch D and the unknown anomaly detection branch S are merged The output of the image realizes the detection of the pixel-level segmentation of the abnormal region of the image. The method disclosed by the invention can improve the segmentation fineness, real-time performance and accuracy of anomaly detection.

Description

technical field [0001] The invention relates to a surface abnormality detection method, in particular to a multi-task detection method for pixel-level segmentation of surface abnormality areas. Background technique [0002] Surface anomaly detection has huge application value in industrial production, medical diagnosis, urban construction management and other fields. Such as the detection of metal surface scratches in industrial production, the detection of wafer surface defects in semiconductor production, the detection of cloth surface stains in textile production, the detection of garbage dumping in urban construction, the detection of surface structure problems of microfluidic chips for biochemistry , detection of brain tumors in medical diagnosis, etc. An effective surface anomaly detection method is of great value for reducing manpower burden, improving product yield, and reducing detection costs. [0003] The existing methods for surface anomaly detection include vi...

AI Technical Summary

Problems solved by technology

[0005] The amount of image data is large, the labeling is highly professional, and the cost of labeling is too high, especially pixel-level labeling

Most supervised methods do not perform well with insufficient training samples and lack the ability to detect unknown anomalies

[0006] Image-based anomaly detection mostly builds different models for small pixel blocks, and multi-scale information relies on the fusion of features collected from multiple image blocks of different scales to obtain it. There are difficulties in utilizing multi-scale features, poor segmentation and positioning of pixel blocks, and poor real-time performance. question

However, using the semantic segmentation network directly through transfer learning has the problems of not being very specific in extracting features and poor in distinguishing ability, and it is easy to misidentify regions with a small proportion as abnormalities.

Images