Design of a Precision Pesticide Spraying System for Smart Agriculture

Abstract

The design of a precision pesticide spraying system that combines explainable artificial intelligence, computer vision, and Internet of Things (IoT)-based robotic control for smart agriculture is presented in this study. A NASNetMobile-based convolutional neural network (CNN) for disease classification, Gradient-weighted Class Activation Mapping (Grad-CAM) for interpretability, a severity estimation stage, and a regression model that forecasts the necessary pesticide quantity are used to analyze crop images obtained by an onboard camera. Robot movement is remotely controlled through an ESP32-based IoT layer via a web/mobile interface, while dosage is administered using a three-nozzle mechanism that exclusively activates the nozzle or nozzles corresponding to the localized infected region. Presented are the hardware architecture, mechanical platform, and full operational process. The AI pipeline and three-nozzle actuation are detailed as the suggested approach for further implementation; dataset preparation, model training, and field validation are scheduled for later phases. This paper presents the results of the first project review phase. The technology is designed to increase crop treatment efficiency and precision while lowering pesticide waste, environmental pollution, and farmer exposure to dangerous chemicals.