Classification system study of soybean leaf disease based on deep learning

Accurate and automatic identification of leaf diseases can help farmers to formulate early response actions to reduce economic losses. A deep learning-based classification system against soybean leaf disease was proposed in this paper. The dataset was derived from multi-blade real crop images taken directly by small UAVs at different time periods and under different illumination conditions. First, for the collected leaf images, SLIC method was used to segment the image of plant leaves by classification system. Then, fine-tuning and transfer learning strategies was adopted to expand training of deep neural networks, as well as data enlargement and Dropout techniques were used to avoid over-fitting. Finally, test results of real data sets are presented quantitatively and qualitatively by means of visualization. The results showed that the classification accuracy can be as high as 99.04% when using Inception-v3 combined with 75% fine-tune parameter strategy.