Real-time identification of plant diseases using aerial robots and deep learning techniques

Agriculture is of paramount importance in providing sustenance to human populations globally. It is a crucial economic growth driver, particularly in emerging nations like India, where majority of the people depend on it. However, the proliferation of plant diseases and pests presents significant threats, leading to substantial declines in crop productivity and agricultural losses. In this regard, real-time identification of plant-based diseases and pests can revolutionise plant monitoring, leading to more efficient and targeted interventions. Hence, in this current research, we present the methodology for performing real-time plant disease detection using lighter versions of deep learning algorithms implemented on sample payloads of mobile robots. Unlike literature that utilises high-end sensors for image acquisition, this study uses off-the-shelf sensors to build an embedded platform and tests the proposed methodology in real-time field surveys. Field studies demonstrate that such a fabricated platform is able to characterise not only the disease but also identify its associated stages with a Mean Average Precision (mAP) of approximate to 96% within 6000 training iterations, achieving a detection speed of 6-10 FPS. We claim using such mobile platform setups can help in effective crop surveillance and timely pest management.