An Integrated System of IoT-based Agriculture, Disease Detection and News Classification

Abstract

Smart agriculture systems are becoming more and more integrated based on IoT and artificial intelligence to aid 
farm monitoring and decision-making, but most current solutions support the assessment of single elements and are studied in laboratory settings, restricting their practical implementation. The current paper introduces a highly integrated, end-to-end smart farming system which incorporates the real-time sensing of the physical environment, automated irrigation management, AI-based detection of plant diseases, and provision of information to farmers into a single, low-cost system. The system is clearly aimed at managing the practical deployment limitations such as
latency, energy efficiency, reliability of connectivity and scalability in agricultural settings with resource limitation. In order to accomplish the deployment-oriented challenges, a comparative analysis of the convolutional neural networks namely ResNet 50, DenseNet-201 and MobileNet-V2 are performed using a large scale data set. DenseNet-201 is more accurate however, MobileNet-V2 is chosen to deploy on the field due to its much low inference time and model size, which is more applicable in limited resources of agriculture facilities. In addition, an agricultural news classification module is incorporated to filter domain specific information for farmers, with limitations in dataset size
and generalization explicitly discussed. Experimental data prove that the proposed system effectively balances sensing, computation and automation requirements, bridging the gap between laboratory-level performance and practical smart agriculture deployment.