Design of image processing anomaly detection algorithm based on cascade convolutional and neural network
Jihai Lei · The Imaging Science Journal · 2025
Aiming at the problem that traditional image anomaly detection methods are unable to detect objects with small or fuzzy targets, the study proposes to integrate residual network-50 into cascade convolutional neural network for algorithm design, and to use residual network-50 to solve the problem of gradient vanishing in the deep network. The results show that the improved algorithm has an area under the image level curve of 0.9546, an area under the precision recall curve of 0.9549, and a false alarm rate of 1.6%, and can effectively detect all targets in the unsupervised defect detection dataset. The algorithm proposed in the study has good practical application effect and reliability, and can perform image anomaly detection with high accuracy. It can reduce the false alarm rate and omission rate in the process of image anomaly detection and provide technical support for industrial production defect detection and safety monitoring.