DEVELOPMENT OF A BIRD DETECTION AND DRONE BASED DETTERENT SYSTEM

Abstract

Bird damage to paddy crops represents one of the most persistent and economically significant biotic stressors affecting agricultural productivity in Kerala, particularly during the heading and grain filling stages. The Baya Weaver bird (Ploceus philippinus), locally known as Attakili, is among the most destructive avian pests in the paddy growing regions. Conventional bird scaring methods including scarecrows, reflective tapes, gas cannons, and manual patrolling suffer from rapid habituation declining rural agricultural labour availability. This study presents the prototype development and evaluation of bird deterrent system, an integrated nano drone based bird deterrent system designed for small and marginal paddy farmers in Kerala. The system employs a two station architecture comprising a ground station and a drone mounted deterrent station. The ground station utilises an ESP32-CAM module equipped with an OV3660 image sensor for continuous field surveillance. Bird detection is performed using the Gaussian Mixture Model (GMM) algorithm,Upon detection, the system simultaneously dispatches a real time alert to the farmer's mobile phone via the Telegram notification platform and, after a ten second delay, wirelessly transmits a 433 MHz RF signal to the drone mounted receiver unit.The drone unit, comprising an Arduino Nano microcontroller and a piezoelectric buzzer, activatesan acoustic deterrent upon receiving the signal. Comparative evaluation of GMM and K-Nearest Neighbours (KNN) background subtraction algorithms was performed across eleven probability threshold values (0.30 to 0.99) using a ground truth dataset of 30 bird occurrences. The GMM achieved a consistent True Positive Rate of 0.900 across all thresholds, with the False Positive Rate declining monotonically from 0.739 at threshold0.30 to 0.423 at threshold 0.99. The KNN model produced a fixed True Positive Rate of 0.767 and an invariant False Positive Rate of 0.906 across all thresholds, indicating threshold insensitivity and substantially higher false alarm generation. GMM operated at threshold 0.99 was selected as the optimum detection configuration. Field deterrence trials over six observation sessions demonstrated that bird deterrent system achieved a mean bird return time of 17.00 ± 2.00 minutes,compared to 2.99 ± 0.52 minutes for conventional manual scaring a 5.7-fold improvement in deterrence duration. The bird deterrent system demonstrates that low cost embedded IoT components, combined with motion based computer vision and aerial acoustic deterrence, can deliver effective, scalable, and farmer accessible bird management for paddy cultivation in Kerala.