Affiliations: [a] College of Computer Engineering and Sciences, Prince Mohammad bin Fahd University, Saudi Arabia
| [b] Dean – Research (Assistant) & School of EEE, VIT Bhopal University, India
| [c] Cloud Computing Consultant, MIT Square, UK
| [d] College of Applied and Supporting Studies, King Fahd University of Petroleum and Minerals, Saudi Arabia
Correspondence:
[*]
Corresponding author. Ghazanfar Latif, College of Computer Engineering and Sciences, Prince Mohammad bin Fahd University, Saudi Arabia. E-mail: glatif@pmu.edu.sa.
Abstract: The agriculture industry is of great importance in many countries and plays a considerable role in the national budget. Also, there is an increased interest in plantation and its effect on the environment. With vast areas suitable for farming, countries are always encouraging farmers through various programs to increase national farming production. However, the vast areas and large farms make it difficult for farmers and workers to continually monitor these broad areas to protect the plants from diseases and various weather conditions. A new concept dubbed Precision Farming has recently surfaced in which the latest technologies play an integral role in the farming process. In this paper, we propose a SMART Drone system equipped with high precision cameras, high computing power with proposed image processing methodologies, and connectivity for precision farming. The SMART system will automatically monitor vast farming areas with precision, identify infected plants, decide on the chemical and exact amount to spray. Besides, the system is connected to the cloud server for sending the images so that the cloud system can generate reports, including prediction on crop yield. The system is equipped with a user-friendly Human Computer Interface (HCI) for communication with the farm base. This multidrone system can process vast areas of farmland daily. The Image processing technique proposed in this paper is a modified ResNet architecture. The system is compared with deep CNN architecture and other machine learning based systems. The ResNet architecture achieves the highest average accuracy of 99.78% on a dataset consisting of 70,295 leaf images for 26 different diseases of 14 plants. The results obtained were compared with the CNN results applied in this paper and other similar techniques in previous literature. The comparisons indicate that the proposed ResNet architecture performs better compared to other similar techniques.
