Plant Disease Detection using CNN and Transfer Learning

The agricultural sector has served as the backbone of The Indian economy for centuries and over the years this area has witnessed advances and growth. According to the current scenario, plant diseases is one of the main hassle affecting the modern agricultural production pattern. For assessing, analyzing and predicting the effects of plant diseases on crop production, the disease severity index is one of the key metrics. Based on the disease severity value the necessary treatment for plant disease may be recommended which is vital in reducing further yield loss. Most of the traditional approach of assessing the plant disease severity involves a trained expert visually inspecting the plant tissue or specimens, but this process turns out to be time-consuming, expensive, and less efficient. So, it’s the right time to come up with disease evaluation systems that could be useful for modern agricultural production. The boom in ML, dissemination of high-quality cameras in mobile devices, and advancements in the field of computer vision allow us to digitize disease assessment techniques. In this paper we have analyzed a multi-class classification of various plant disease over 2 different datasets: PlantVillage and PlantDoc, by using various pre-trained models like VGG16, Xception, InceptionV3, ResNet152, MobileNetV2. We have also analyzed the drawbacks and effectiveness of each dataset and its producibility for using them in real-time detection systems which can be beneficial in reducing losses that we face because of disease outbreaks.