PLDD—A Deep Learning-Based Plant Leaf Disease Detection

Abstract

Plant diseases have a detrimental effect on the health of food production in agriculture. As a consequence, it significantly decreases the quality, quantity, and productivity of the yield. Thus, for sustainable agriculture, automated detection and diagnosis of plant diseases at an early stage of growth are highly desired. While several computer vision based applications have been suggested for this process, they still suffer from long-lasting training/testing time with large datasets. Besides, due to the hardware limitation and computational complexity, such model development is crucial in handheld devices. This research presents a new transfer learning-based optimized EfficientDet deep learning framework as a practical solution for automated plant disease detection. A total of 3,038 images are collected from two public dataset repositories and annotated manually to train the model. The proposed model performance is evaluated in terms of mean Average Precision (mAP). It achieves an overall mean average precision. of 74.10% with a substantially fewer number of parameters and FLOPs than other state-of-the-art approaches. Such a framework can be implemented on devices with minimal computing resources to make reliable and timely decisions.