Intelligent understanding algorithm of aerial visual inspection image based on deep learning technology

Abstract

Abstract At present, there is a problem that the efficiency and accuracy of the safety belt wearing detection method for aerial patrol workers are not ideal. In order to solve this problem, based on the deep learning technology, a safety belt specification wearing detection model is constructed. First, in view of the low quality of image data, wavelet transform and Gaussian curvature filter are used to preprocess the image. Aiming at the defect of poor performance of convolutional neural network (CNN), Gabor local features and Momentum algorithm are used to improve it. Finally, combined with the above content, a safety belt specification wearing detection model based on improved CNN is constructed. The results show that the loss value of the model is 0.51, the accuracy rate is 98.14%, the Recall value is 95.04%, and the AUC value is 0.971. Therefore, the model built in the study can detect the wearing of safety belt with high efficiency and accuracy, and ensure the safety of staff.