Research on Edge Detection Operator of a Convolutional Neural Network

Abstract

With the rapid development of image technology, edge detection technology is becoming more and more diverse. In order to fully improve the edge detection effect, it is necessary to innovate in the detection algorithm. In recent years, deep learning has become an emerging research direction in the field of machine learning, and convolutional neural network[1] (CNN) is a deep learning model that has been widely studied and applied. Since CNN can automatically detect the multi-directionality of image features, it is widely used in image classification and recognition. The traditional edge detection operators are based on gradients, which can only be detected from several fixed directions, resulting in the loss of edge information in other directions. For the above, edge detection algorithm is proposed based on CNN in this paper. The algorithm is as follows: Firstly, because the convolutional network can automatically extract image features. In this paper, CNN is constructed under the framework of tensorflow to obtain the edge detection operator after training. The prepared data set is input into the network for training, and the training result is saved on matlab for verification. In addition, two comparison experiments were carried out in the experiment. The first set of experiments uses six different sets of optimizers to train the network and generate corresponding edge detection operators. They are used to perform edge processing on the image of the test set. By comparing the result of processing, the optimizer suitable for this paper was selected as Adam Optimizer and Gradient Descent Optimizer. The second set of experiments was performed by comparing the edge detection operator generated by Adam Optimizer with the classical operator. The experimental results show that the edge detection operator based on convolutional neural network detects more abundant edge information than the classical operator, and also achieves higher effect on the edge detection of the target.