Recognition Algorithm Based on Convolution Neural Network for the Mechanical Parts

For the problems that the traditional mechanical parts identification algorithm needs to design and extract relevant features artificially, so that the process is complex and time consuming in the computation is larger as well as identification accuracy is easily affected by the diversity of parts morphology, a mechanical part identification algorithm based on convolutional neural network is proposed in this paper. The Leaky ReLU function algorithm as an activation function is used to improve the pooling method, and a SVM classifier is combined to construct a convolutional neural network WorkNet-2 for the recognition of mechanical parts. In the recognition experiments of common four kinds of mechanical parts, the trained WorkNet-2 network's recognition accuracy on the test set reached 97.82%. The experimental results show that compared with the traditional mechanical parts recognition algorithm, this algorithm can extract the high-level features of the target parts, and has the advantages of small influence of parts shape diversity, the recognition rate is higher and good realtime performance.