Robotics Technology and Computer Vision for Precision Agriculture

Automated robots and computer vision-based inspection of plant health provide a revolutionary solution to the field of agriculture. Such a solution can address the emergent challenges of diminishing crop yields, declining soil quality, and climate change. In addition, it can mitigate the negative impact of labor market issues. Hence, vision-based robots can contribute to improving biodiversity and better sustainable agriculture and ecosystems. In this study, a computer vision-based framework for plant disease identification is presented. The framework comprises two stages of augmentation and classification. Extensive data augmentation is applied to provide a range of cases that simulate real-life conditions. The impact of gamma correction, blurring, cropping and noise augmentations is investigated to determine the robustness of the classification. A lightweight Convolution Neural Network (CNN) is introduced for multi-class classification. An analysis of its performance together with state of the art architectures is detailed. Competitive performance in terms of accuracy, precision, recall, and F1-score is demonstrated by the introduced CNN. The proposed CNN offers a testing accuracy of 95.9%. The main improvement provided by the architecture is the reduced average time per epoch for training and inference times with a reduction of 57.9083% and 31.609% respectively.