Tailored Deep Learning based Architecture for Smart Agriculture

Abstract

Disease detection in a plant or tree using traditional ways such as the farmers expert naked eyes is both time and resource consuming and may engender tremendous crop losses. Thus, the early diagnosis and treatment of these diseases can minimize the losses in the whole crop and can improve quality and diversity for the consumer later. With the recent advances in Deep Learning, powerful approaches are developed for both detection and classification that can cope with complex environments. In this paper, we propose an efficient deep learning-based architecture for object detection in the context of Smart Agriculture. The proposed solution combines deep learning and tweaked transfer learning models for object detection with balanced data for every class of images. It can operate in a more complex environment and takes into consideration the state of the input. Its aim is to automatically detect damages in leaves and fruits, locate them, classify their severity levels, and visualize them by contouring their exact locations. Numerical results reveal that the proposed solution, based on Mask-RCNN achieves higher performances in features extraction and damage detection/localization compared to other pre-trained models such as VGG16 and VGG19.