VISUAL RECOGNITION OF NOISY FASTENING BOLTS USING NEURAL NETWORKS AND WAVELET TRANSFORM

摘要

The paper focuses on the problem of automatic detection of the absence of the fastening bolts that secure the rails to the sleepers. We have developed a vision based system that combines a pre-processing technique based on, Wavelet Transform (WT) and two neural network architectures, Multilayer Perceptron Network (MLPN) trained by the Back-Propagation algorithm and Radial Basis Function Network (RBFN). Furthermore we have tested the classification performances of the networks in presence of gaussian noise on the image. The high percentages of obtained detection rate using real images show a high reliability and robustness of the system even.