Implementation of a Face Detection Surveillance Security System in a Rice Farm Using Supervised Machine Learning Techniques

Abstract

This paper presents an innovative approach to developing a surveillance security system tailored for the agricultural sector, specifically rice paddy fields in Thailand, by employing supervised machine-learning-based face detection techniques. The proposed system integrates real-time facial recognition with automated notifications sent to the owner via text messages and facial images through the LINE application. Video inputs from on-site CCTV cameras are processed into a sequence of images, from which key facial features are extracted using the Haar-like feature extraction method. Identification of individuals as either authorized owners or intruders is achieved through supervised machine learning utilizing the K-Nearest Neighbors (K-NN) algorithm. Upon detecting an intruder, the system promptly activates an alarm and sends notifications, enhancing real-time security monitoring. The system is implemented using a Raspberry Pi microprocessor and is powered by a stand-alone solar energy system, ensuring sustainability and operational efficiency. Performance evaluation includes extensive testing, verification, and comparison against conventional methods using the widely recognized Labeled Faces in the Wild (LFW) dataset. Experimental results demonstrate that the proposed system achieves an average accuracy of 86.86%, slightly surpassing traditional techniques. Additionally, it exhibits a significantly improved recognition speed, averaging 8.3 seconds per detection. Robustness and adaptability were further assessed by evaluating the system under varying brightness and distance conditions. The findings confirm the system's ability to provide real-time, precise facial recognition of intruders, thereby establishing its effectiveness as an advanced surveillance security solution. The system's operational boundaries for successful intruder recognition and warning alerts are set within a 1 to 3-meter range from the camera. Recognition accuracy was observed at 95% under optimal conditions and 83% in more challenging scenarios, while the warning alert system demonstrated a success rate between 86% and 97%. These results highlight the proposed system's superior performance and reliability, making it a valuable security solution for agricultural applications.