On the Use of Class Activation Map on Rice Blast Disease Identification and Localization

The process of disease evaluation requires an expertise to examine through all the experiment rice subjects individually. This problem produces a lot of workload and could be a great barrier in the rice breeding research areas. In this study, we proposed an automatic rice blast disease identification and localization based on a deep learning method where the final goal is to identify blast disease image whether it is a healthy, moderate or susceptible disease. Thus, we need to be able to localize the disease areas over the rice image. To achieve this goal, the proposed method employed the Class Activation Map (CAM) on the U-NET model using convolutional neural networks (CNN) where the network was trained with only two alternative labels, absence and presence of disease. With the class activation maps, we are able to generate the heatmap which also allows the disease areas to be weakly localized.