Experimental study on crop disease detection based on deep learning

• Published in: IOP conference series. Materials Science and Engineering 2019-07, Vol.569 (5), p.52034
• Main Authors: Xia, Xiaoling, Wu, Yongbo, Lu, Qinyang, Fan, Changqi
• Format: Article
• Language: English
• Subjects: Accuracy; Artificial neural networks; Crop diseases; Datasets; Deep learning; Image classification; Machine learning; Object recognition; Parameters; Plant diseases; Training
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/569/5/052034

At present, deep learning has been widely used in our daily life, and the recognition of crop disease degree based on deep learning has gradually entered the public's vision. The degree of crop diseases can usually be judged by the characteristics of leaf colour, shape and spot. According to the data set provided by AI Challenge Competition, the degree of crop diseases can be divided into many situations. According to the current open source data, the recognition accuracy rate on this data set has not been able to exceed 90%. In order to achieve better recognition results, on the basis of the convolutional neural network CNN, the transfer learning method is used to load the residual network ResNet pre-training model and SeNet + ResNet to carry out deeper network training to improve the efficiency and accuracy of recognition. This paper carries out many experiments under the condition of controlling variables. The experimental results show that Resnet can play a good role in this data set with training parameters and achieve 83% accuracy, while SeNet + ResNet can achieve 90% accuracy without using pre-training parameters, and 87% accuracy of test set recognition. It can be seen that the SeNet + ResNet method has better effects in image classification and recognition tasks in this data set.