Research on Image Object Fault Detection Based on Deep Learning

In recent years, with the rapid development of deep learning, many computer vision tasks have achieved breakthroughs again and again. In many computer vision tasks, fault detection has been widely concerned because of its great application value. Compared with traditional image target fault detection methods, the image target fault detection method based on deep learning digs image features layer by layer through deep neural networks, and can learn more advanced abstract features without manual feature design, and has better generalization. Based on deep learning technology, this paper studies the algorithm of image target fault detection. In this paper, a fault detection method based on attention mechanism is proposed to solve the problem that the fault detection based on object detection cannot distinguish the fault with not obvious features. This method uses the attention mechanism to assign different weights to each feature graph when the network performs multi-scale feature fusion, so that the network selectively pays attention to the higher-order semantic information and lower-order detail information that are more helpful for fault identification, and improves the recognition ability of the network for the faults with less obvious features. In the experimental verification stage, the performance of the proposed method is higher than that of the benchmark method on both private and public datasets. Among them, 93.28% mAP@all is achieved on the private dataset, which is 1.16% higher than that of the benchmark method (92.04%). 78.55% [email protected] was achieved on the publicly available VOC dataset, an improvement of 1.48% compared to 77.07% for the benchmark method. At the same time, FPS of this method can be maintained at 74,[email protected] up to 99.55%, which can fully meet the real-time requirements and achieve high detection accuracy, which proves that the algorithm in this paper has reached the advanced level.