An Explainable AI-Driven Hybrid Deep Learning Model Integrating Lightweight CNN And Neural Architecture Search Network For Precision Agriculture Applications

Crop productivity is most important in the economic growth and development of the world. Unfortunately, productivity is on the decline while the population keeps growing, making it more challenging to produce food globally. There are several factors that contribute to the decaying productivity, including desertification, crop crises, among others. Here in this paper, we concentrate on crop crises due to plant diseases. These illnesses have a significant effect on crop yields, but they can be controlled and managed using conventional practices and AI-based contemporary techniques. Conventional practices are time-consuming and cannot always ensure the best outcomes, while AI-based methods like deep learning can yield more precise and timely results. Various pre-trained models like Alex-Net, VGGNet, Inception-Net, and ResNet have proved to be very accurate—usually over 90%—when it comes to classification problems. Yet these models are computationally expensive and require high memory and computational resources, which is resource-intensive in some environments. In order to solve these problems, this paper introduces a hybrid model, drawing on principles of Explainable AI in an attempt to provide optimized outcomes cheaply and within a reasonable timeframe. XAI has the major advantage over simply using pre-trained models by increasing the transparency and trustworthiness of the decision-making process. XAI techniques allow researchers and farmers to see what areas of the potato leaf the model is looking at when making a prediction of a disease and aid in checking if the model is learning actual patterns of diseases or getting misled by irrelevant features such as background noise. This insight aids in better model tuning, increases user trust, and leads to more reliable and helpful disease control in real farming conditions.