Lightweight deep CNN models for identifying drought stressed plant

Drought is one of the most severe climatological disasters that has negative impact on agricultural production around world. Over years, computer vision technology been used in conjunction with machine learning applications to replace traditional destructive and time-consuming methods for real-time monitoring drought-affected plant. Deep (DL) techniques have gained a stellar reputation image classification recently, convolutional neural network (CNN) emerging as industry standard. However, size deep CNN models frequently large due massive number parameters field application often not feasible limited storage computational resources. Several lightweight selected based less than 6M were trained tested. The EfficientNet model achieved accuracy 88.12 88.97 percent identifying drought, mild no drought plants visible near-infrared images respectively. findings this study can be assist development automated early detection stressed plant sizes suitable diagnosis mobile or embedded devices.