An Approach for Improving Accuracy and Optimizing Resource Usage for Violence Detection in Surveillance Cameras in IoT systems

Abstract

Smart farming that uses information and communication technology is developed as a critical technique to address the challenges related to agricultural production, environmental effects (climate change), food security, and supply chain. The recent statistics reveal that the world's population has been increased significantly, which is expected to reach 7.7 billion. It is essential to achieve a significant rise in food output to meet the requirement of such a massive growth of population. However, due to the natural conditions and a variety of plant illnesses, food productivity and farms are reduced. In order to diagnose food diseases in farming, new technologies like the Internet of Things and artificial intelligence are now essential. To this end, the research paper introduces a novel artificial intelligence model represented by a twelvelayer deep convolution neural network to identify and classify plant image diseases. 38 distinct types of plant leaf photos are used for training and testing the suggested model, which are obtained by adjusting different parameters such as (a) hyperparameters; (b) coevolutionary layers; (c) and pooling layers in number. The proposed model consists of an extractor and classifier of functions. The first section involves three phases, i.e., it consists of two convolution layers and a maximum pooling layer for each phase. The second section consists of three levels: flattening, hidden, and output layers. The proposed model is compared with LeNet, VGG16, AlexNet, and Inception v3, which are considered state-of-the-art pre-trained models. The results demonstrate that the accuracy of LeNet, VGG16, AlexNet, and Inception v3 is given as 89%, 93%, 96.11%, and 97.6%, respectively. The findings provided in this research show that the suggested model outperforms state-of-the-art models in terms of training speed and computing time. Also, the results show that the proposed model achieves a considerable improvement in terms of accuracy and the mean square error compared to the state-of-the-art methods. In particular, The outcomes demonstrate that the suggested model achieves a mean square error and prediction accuracy of 98.76% and 0.0580, respectively. The results also depict that the proposed model is more reliable, allows fast convergence time in obtaining the results, and requires only a small number of trained parameters to identify the plant diseases accurately.