NEXT-GEN AGRICULTURAL SECURITY: INTELLIGENT MULTI-SENSOR SURVEILLANCE WITH DEEP LEARNING-BASED BACKGROUND SUBTRACTION AND CNN CLASSIFICATION.

Abstract

Large farmers are now begin to apply the new technologies to control their farms. In the new technologies, CCTV cameras are mainly used. These CCTV cameras are used to monitor the farm field manually. At present, some large farmers are using IoT-based systems for full control of the farm field. This system is also capable of identifying stray animals. Such systems are not affordable to any of the small and marginal farmers. This work designs and develops a low-cost intelligent farm field surveillance system to prevent the farm field from stray animals and stealing of farming equipment. A deep learning CNN model is proposed and trained in this work for the classification of objects into animals, suspicious human activities, non-suspicious human activities, and usual motion activities on farms. The proposed CNN model is trained and tested on the specifically developed dataset. The proposed CNN model achieves 92\% average accuracy in the classification of activities with 0.92 precision and 0.91 recall. The proposed CNN model is also compared with well-known deep learning models i.e.MobileNet, VGG16, and ResNet50. The comparison of such models is carried out on the same dataset. The obtained results show that the proposed model outperformed the existing deep learning model in accuracy and precision. The accuracy of the proposed work is 8% higher than the ResNet, 6% higher than the VGG16, and 31% higher than the MobileNet. The obtained results establish the correctness and consistent performance of the proposed model.