# An AI-powered smart Agribot for detecting locusts in farmlands using IoT and deep learning

## TL;DR

This paper introduces an AI-powered Agribot that uses IoT and deep learning to detect locusts in farmlands with high accuracy.

## Contribution

The novel contribution is an Agribot system combining IoT, deep learning, and feature selection for efficient locust detection in real field conditions.

## Key findings

- The Agribot achieved 99.51% accuracy in locust detection using VGG19 and Logistic Regression with the SVC feature selector.
- The Agribot operated efficiently at a speed of 0.3048 m/s with live video streaming in agricultural fields.
- The system received a System Usability Scale (SUS) score of 86%, indicating strong user satisfaction.

## Abstract

In many countries, locusts have significantly harmed agricultural production. To prevent their spread, the Agriculture Robot (Agribot) with cutting-edge technologies like the Internet of Things (IoT) and Machine Learning (ML) can be a possible solution. Thus, this study presents an astute way to develop an Agribot using IoT, ML, and DL-based architecture for detecting locusts in agricultural fields. The IoT framework ensures proper automation by utilizing various agriculture-related sensors, a centralized Android application, and an IoT cloud server. In contrast, the ML and DL methods include several pre-trained Convolutional Neural Network (CNN) models with conventional ML classifiers and a nature-inspired algorithm such as Artificial Bee Colony (ABC) and the SVC feature selector. To assess the proposed system’s efficacy, experimental data have been collected and interpreted accordingly. This research achieved the highest accuracy of 99.51% in locust detection using the VGG19 pre-trained CNN model with Logistic Regression (LR) and the SVC feature selector. In addition, the Agribot operated efficiently at a satisfactory speed in agricultural fields with live video streaming. The maximum speed of the Agribot was recorded at 0.3048 m/s. Furthermore, the study obtained a SUS score of 86% for the developed system. Although the system performs well in locust detection and automation in real field conditions, the research also identified some limitations during the study and implementation. However, the developed application demonstrates strong feasibility for real-time locust detection in agricultural fields.

The online version contains supplementary material available at 10.1038/s41598-025-23497-8.