Enhanced CNN and SVM with Adaptive Modality Switching and Audio-Based Video Summarization for Real-Time Agricultural Intrusion Detection

Smart intrusion detection in agriculture involves the use of IoT, AI, and sensor-based technologies to monitor fields for unauthorized human and animal activity. Advanced AI models enhance detection accuracy, reducing false alarms and improving response efficiency. The integration of edge computing and cloud-based analytics ensures rapid data processing, making intrusion detection systems more effective and reliable in modern agricultural security. Traditional security systems rely on either video-only or audio-only detection, and struggle in low-light conditions due to the absence of adaptive switching mechanisms. They typically lack event-based video summarization, leading to prolonged, redundant footage storage. Additionally, these systems are high-end and cloud-based, requiring significant computational resources. They are generally designed for broad security applications rather than specialized use cases. The proposed agriculture intrusion detection system consists of an audio based video summarization(ABVS) intrusion detection using an external sensor and day light conditions. The audio based video intrusion system consists of hybrid model with a yolov4-tiny model integrated to a machine learning based audio model which uses MFCCs, Delta and Delta-Delta MFCCs, Chroma features, SMOTE, SpecAugment techniques to improve detection accuracy. An integrated convolution neural network (CNN) model and an integrated SVM model based system termed as “audio-video intrusion detection system with adaptive switching” based on light conditions was implemented with accuracy of 98% and 94%. The comparative t-SNE plot insights were logically used to improve the accuracy of the model as well as the noise augmentation technique with MFCC plus spectral features assisted in achieving the highest efficiency. Keywords: Precision Agriculture, Sensor Network, Soil Monitoring, Weather Monitoring, Real-Time Data, Agricultural Productivity, Sustainability.