关键词: 深度学习, 卷积神经网络, 图像分类, 图像检测

Abstract: At present, the convolutional neural network（CNN） model must be retrained when the detection target is replaced, which makes the replacement of the detection target take more time, the training cost increases, and the understanding of the model also improves. To solve this problem, the method of image classification using CNN is proposed. Firstly, each detection state of the detection target is classified. Then, the image classification model of CNN is used to classify the input image in real time. The experimental results show that this method can quickly change the detection target, the detection accuracy can be improved to 99.9 %, and the training cost and personnel technical requirements are also greatly reduced.

Key words: deep-learning, convolutionalneuralnetworks, imagedetection, imageclassification