Identification of Plant Leaf Diseases Based on Inception V3 Transfer Learning and Fine-Tuning

Abstract

Crop disease is a major factor currently to jeopardize agricultural production activities. In recent years, with the great success of deep learning technology in the field of image classification and image recognition, and with the convenient acquisition of crop leaf images, it is possible to automatically identify crop disease through deep learning based on plant leaf disease images. This paper mainly completed the research and analysis of leaf disease identification of agricultural plants based on Inception-V3 neural network model transfer learning and fine-tuning. A large number of model accuracy tests are carried out by training neural networks with different parameters. When the network parameter Batch is set to 100 and the learning rate is set to 0.01, the training precision and test precision of the network reach the maximum. Its training precision rate for crop disease image recognition in the PlantVillage DataSet is 95.8%, and the precision rate on the test set is as high as 93%, and far exceeding the accuracy of manual recognition. This fully proves that the deep learning model based on Inception-V3 neural network can effectively distinguish crop disease.