TRANSFER LEARNING-BASED SMART CROP PROTECTION SYSTEM FOR ANIMAL DETECTION AND DETERRENCE

Abstract

Crop depredation by wild and stray animals continues to be a perennial problem in agriculture, resulting in enormous losses in productivity and pressure on the economy of farmers. To overcome this, in this paper, we present a Transfer Learning-Based Smart Crop Protection System based on the YOLO (You Only Look Once) object detection model, for real-time detection and prevention of animal activities. Through the transfer learning of a pre-trained YOLO architecture, the system is trained to detect certain types of animals causing crop damage, even with a small amount of domain-related images. Live video streams obtained from field-mounted cameras are analyzed to estimate the presence of animals with high accuracy and low vibrant. Once detected, the system can automatically initiate nonlethal countermeasures, such as sound alarms or flashing lights tailored to the type of animal identified. The Internet of Things(IoT) and edge computing are incorporated, which facilitates on-site computing without always being connected to the cloud. Experimental results suggest that our system can accurately detect elastic object approximations under different lighting and environment situations, with strong deterrent responses and decreased false positives. The system offers an intelligent and automatic scale value-added crop protection by effectively employing transfer learning and deep learning models. The accuracy of the proposed model in detecting the animal is 93%. This Artificial Intelligence (AI) approach saves the crop from animals.