An Improved Mobilenetv2 for Rust Detection of Angle Steel Tower Bolts Based on Small Sample Transfer Learning

Rust detection of diagonal steel tower bolts is dangerous and heavy, so it needs to be assisted by intelligent technology. In order to be applied in the equipment with limited resources such as mobile terminal or embedded terminal, based on the method of deep learning, this paper proposes a new lightweight neural network to solve the classification problem of rusty or non-rusty angle steel tower bolts. The network is improved on the basis of mobilenetv2. In order to be suitable for the training environment with small samples, the strategy of transfer learning is used. In order to make up for the loss of characteristic information caused by convolution, a double branch structure is used to better obtain characteristic information. Using the attention mechanism, the network can automatically obtain the important weight of features according to the training process, so as to better improve the separation performance. In the task of bolt rust detection, the proposed network obtains an accuracy of 97.6% and an F1 score of 96.3% while ensuring small parameters. Compared with other excellent neural networks, the proposed method has better classification performance and is suitable for outdoor environment detection.