An Agricultural Unmanned Ground Vehicle for Localized Spraying of Harmful Plants Using Convolutional Neural Networks

Abstract

The utilization of artificial intelligence (AI) in agriculture through unmanned ground vehicles (UGVs) has the potential to significantly reduce excessive chemical use and environmental pollution. Instead of spraying entire areas, local interventions can be carried out with agricultural UGVs, allowing for the removal of unwanted harmful plants by local spraying. This approach leads to a more effective and healthy outcome compared to dispersing pesticides in the air. The harmful chemical effects on agricultural lands are minimized, spraying is carried out more efficiently, and costs are reduced through a decrease in the amount of pesticide used. In this work, we developed an autonomous agricultural ground vehicle with a unique design to locally spray harmful plants in agricultural lands. The vehicle's design was inspired by nature, and aluminum profiles were preferred in its production for lightness. An innovative suspension system was established to facilitate the movement of the vehicle in farmland, minimizing uneven ground conditions. A simultaneous roadmap was generated via the robot operating system using data received from the depth camera placed on the vehicle. A detection algorithm based on convolutional neural networks was used to detect harmful plants in agricultural land. Prompt and effective spraying is crucial once harmful plants are detected, as inefficient spraying may lead to the proliferation of these plants. To overcome this bottleneck, the detection algorithm used enables autonomous spraying with the least delay in real-time. In field tests carried out with the developed ground vehicle, the harmful plant in the agricultural land was detected with an accuracy of 90% from a height of 40 cm and sprayed successfully instantaneously. Critical parameters such as vehicle position, battery consumption, and pesticide status are monitored simultaneously during spraying with the control and monitoring interface developed via an external ground station. The power supply components of the vehicle are determined such that the vehicle can be operated on a relatively flat agricultural land for 1.5 - 2 hours. Additionally, solar cells placed on the vehicle are intended to extend its duty period. Overall, this study demonstrates the potential of AI-assisted UGVs in agriculture for reducing excessive chemical use and environmental pollution while increasing efficiency and cost-effectiveness.