Pplication of deep learning model in recognition of growth stages of Cucumis melo L. in greenhouse

Abstract

Despite the increasing application of deep learning (DL) models in various socioeconomics such as financial analysis and forecast, intelligent transport, self-driving, disease diagnosis, the effective use of this technology to support agricultural cultivation is still limited. This paper introduces the implementation of the lightest and state-of-the-art YOLOv5 architecture for automatic recognising of important growth stages of Cucumis meloL. from the camera images collected in the greenhouse. This image identification initiative achieved an average accuracy of 96% F1-score in the identification of the five growth stages of Cucumis melo L. using a limited set of training and testing data (total 2,818 images of Cucumis melo L.). These preliminary results lead to the conclusion that the YOLOv5 object detection and classification model is a truly lightweight and promising DL solution after the adoption of the transfer learning technique. Moreover, the YOLOv5 model can execute good performance on edge devices which may open up a new approach in different object detection and classification in real-time directly from a smartphone, Jetson Nano, IP camera...