Research On Recogniton Technology Of Micro Parts Based On Convolution Neural Network

In recent years,due to the compact structure,stable performance,low energy consumption and strong anti-interference ability of precision micro electromechanical products obtained by micro assembly,which have been widely used in various fields.In order to improve the recognition accuracy and recognition efficiency of micro parts in micro-assembly tasks,this paper studies the recognition technology of micro-assembly based on convolutional neural network to solve the problem of positioning the micro parts in the microassembly tasks.Firstly,a micro vision acquisition system is built in this paper,the question of image preprocessing is studied before recognitioning micro parts.Aiming at the question that the contrast between part features and image background is not strong,the image of histogram equalization is limited.Aiming at the noise caused by the overheated camera sensor,the image is filtered by Gauss filter.In order to speed up the convergence of neural network in the training process,the input image is normalized.Secondly,the recognition technology based on the convolution neural network of micro parts is studied.Due to the small proportion of parts in the drawing,the deep and shallow features of the network are fused,and the depth separable convolution is used to ensure the recognition efficiency of the model.In order to get the angle information of parts,the region proposal network(RPN)is improved,three scales,three proportions and six angles of recommendation frame are adopted.In order to avoid the error caused by rounding,the sub-pixel position is calculated by bilinear interpolation.In order to avoid the interference among similar parts in the classification of the recommendation box,the center loss function is added in the training to reduce the class spacing.Finally,the training strategy of the network model is studied.After thenetwork is initialized,the model is trained by the transfer learning and end-toend training.Finally,the effectiveness and superiority of the algorithm are verified by the test set.