Image Classification Using Deep Learning Algorithms for Cotton Crop Disease Detection

Agriculture plays an important role in the development of the economy of a nation. Many diseases caused by pathogens and pests hamper agricultural production. If not detected earlier the diseases can cause severe damage, that is why the most important step in the prevention of damage is to detect the disease as early as possible. Traditionally the diseases are detected on basis of past knowledge using bare eyes. The traditional process can be harmful as incorrect detection can account for the wrong and excess use of pesticides harming plants. This paper presents a system for detecting crop diseases using deep learning-based image classification of crop leaves. The system can detect three cotton diseases- Bacterial Blight, Curl Virus, and Fusarium Wilt - by scanning cotton plant leaves. The paper also compares performances of 4 different deep learning architectures. The highest accuracy of the system obtained using ResNet152 V2 architecture is 99.12% on the training dataset and 98.26% on the testing dataset.