Research on image classification algorithm based on DenseNet for small sample in industrial field

Abstract

Image classification is an important branch in the field of computer vision and it has a wide range of application scene. The level of automation has been greatly improved by image classification technology in the field of industry, effectively reducing the complexity of many tasks. However, traditional image classification methods cannot adapt to the complex and changeable environment in practical application of the industrial field. The recognition accuracy is low and the maintenance of these industrial systems is difficult. Therefore, an improved DenseNet network model called small kernel densely convolutional neural network (SK-DCNN) is proposed to improve the performance of small sample image classification in the industrial field. Small convolution kernel in SK-DCNN can fully and effectively extract more subtle features in the image. The dense connection method of dense blocks further improves the reuse of features on the basis of DenseNet. Using a small amount of data to train the model can obtain high classification accuracy. The experimental results of the article show that SK-DCNN achieves good performance (about 96.9%), which is about 4% higher than the accuracy of DenseNet. The comparative analysis with other typical neural network models also shows that the algorithm proposed in this paper has better classification effect in complex small sample images.