Research On Object Detection Method Of Industrial Parts Based On Deep Convolutional Network

With the rapid development of computer vision technology in industry,the past method of relying on manual part sorting is gradually being replaced by machine inspection.The method of using machine to sort parts has certain requirements on the operator and the process used,and the complex factory environment can easily lead to errors in traditional machine sorting.Deep learning object detection methods can automatically extract features from a large number of data samples,which helps to improve the accuracy of object detection.At present,deep learning-based object detection algorithms still have some problems in practical applications.On the one hand,there are problems that materials obscure each other,features are difficult to extract,and targets cannot be precisely located on industrial part identification tasks.On the other hand,many industrial equipments in the field have weak arithmetic power,and the traditional deep network model is prone to generate a large number of computational parameters,which requires equipment with high computational power for computation,thus leading to the traditional deep network model cannot be well applied to some industrial equipments.Therefore,it is still a challenge to compress the deep network model to reduce the model size and speed up the detection rate of the model.In this paper,the two problems of low accuracy of detection methods and limited computing power of industrial equipment faced in the field of industrial part object detection are studied in depth,and the main innovation points are summarized as follows.(1)A dual recognition method based on deep convolutional network and Hough circle transform is proposed for the current problems of target overlap,difficulty in locating the center,and poor target wrapping in the part recognition process in industry.Firstly,the data set is pre-processed to complete the sample expansion;then,based on the deep convolutional network,a spatial pyramid pooling layer is added to the network structure,while the prior frame and loss function are improved according to the specific data set of this paper to complete the extraction of the target prediction frame;finally,the Hough circle transform is added to the target for secondary recognition to achieve the purpose of accurate recognition of the part area.The method is compared with other detection methods on the valve part dataset,and the experimental results show that the method not only improves the detection accuracy of the model,but also enhances the wrapping degree of the target,which is of high value in practical applications.(2)To address the problem that current deep learning-based object detection methods are difficult to operate in industrial equipment with low computing power and limited resources,a part image object detection method based on a lightweight deep convolutional network is proposed.Firstly,the ordinary convolution in the network is replaced by Ghost convolution,which effectively reduces the number of convolution kernels;secondly,the new G-ResNet residual structure is generated by optimizing the residual network to complete the intra-layer fusion of the convolutional residual blocks;finally,the reconfiguration of the convolutional approach is realized by drawing on the idea of depth-separable convolution.The experimental results show that this method achieves the lightweight of the model through triple compression processing,saves the cost of device memory,reduces the parameters of the model,decreases the computation of the network,and reduces the size of the model.Compared with the previous uncompressed deep convolutional network,the detection accuracy of the target is somewhat reduced,but it is sufficient to accomplish the task of detecting the target of the part in the industrial process.