Fine-Tuning-Powered Plant Disease Detection Based on Cloud-Edge Computing in Smart Farms

Plants are an indispensable part of agricultural production, and the study of plant disease classification and recognition has an important impact on garden horticulture, environmental protection, crop growth and the continuous development of agricultural economy. Moreover, it plays a very important role in crop production intelligence and crop big data management applications. Traditional machine learning relies too much on expert knowledge, and the form of artificial extraction of plant leaf information cannot meet the recognition of complex diseases. In this paper, we come up with an edge-cloud plant disease classification approach by using fine-tuning to improve the performance continuously. Methods of crop disease identification leverages convolutional neural networks (CNN) that achieves successful performance in various computer vision tasks. The improved leaf recognition model performs the collection, analysis, processing and prediction of heterogeneous data in the process of crop planting. The proposed fine-tunning-improved CNN model is used for searching three common potato diseases in real time. In the cloud, the dataset is deployed to derive the prediction model using CNN for performing fine-tuning in the edge device. Based on the presented approach, the performance of the model is improved and the time consuming decreased by fine-tuning method in the edge environment.