Real-Time Pest Detection System Using Efficientnet Deployed On Raspberry Pi For Precision Agriculture

A real-time pest classification system is developed for this study with the help of Efficient Net architecture for precision as well as for resource efficiency. The hardware setup includes a Raspberry Pi 4 Model B, 4 GB RAM, DC-to-DC Buck Converter (3A voltage regulator), 5MP Raspberry Pi Camera Module, and 16×2 LCD module. The system architecture includes a 12V adapter powering a buck converter that supplies the 5V needed for the Raspberry Pi and a VNC viewer that is used to access and execute the application remotely over a Wi-Fi connection. The model was pre-trained, and the stored model is utilized in the memory card using the transfer learning methods to optimize computation. Images are taken and preprocessed using the Raspberry Pi Camera Module and resized into a 50×50 resolution. However, robust model generalization is enhanced by using data augmentation techniques such as random rotations, flips, and even brightness adjustments. The architecture of the proposed training pipeline uses an EfficientNet model set with a batch size of 32, Adam optimizer, and a learning rate of 0.001. Thus, training is performed for 15 epochs, at the end of which the model attains a classification accuracy of 93.5%The LCD module presents the classification result, and also sends the prediction to the cloud-based server through API. The proposed system is a scalable low-cost solution for precision agriculture that would empower stakeholders in real-time detection of pests while minimizing crop loss. This paper demonstrates the system's effectiveness in a resource-constrained environment, leading to future enhancements including dataset expansion and edge AI deployment for localized decision-making.