A next-generation hybrid generative learning framework integrating dual GAN and Swin Transformer–CNN for robust early diagnosis of pomegranate diseases

Abstract

Nowadays, health is the most important thing, and everything is hybrid, so we need to take care of our health, which is our first priority. To identify pomegranate diseases and determine whether a pomegranate is healthy or dis-eased, this research proposes a novel hybrid method that combines the SWIN transformer and CNN models. It separates healthy from diseased pomegranates using generative AI and a dual-GAN system, thereby enhancing training and out-comes. While CNNs categorize and diagnose illnesses, the first model, the SWIN Transformer, captures complex information from real-time field data. Here, we present methods such as data augmentation, rotation, resizing, and random modifications to improve the model’s performance and enable it to adapt to various field conditions. The results we obtained showed 85.2% accuracy on the training data and 78.6% on the new data. It can differentiate between healthy and dis-eased samples, identify complex patterns, and accurately categorize data. These advantages enable more accurate farming and address the problem of insufficient data. The experiments show that adding more data sources, such as environ-mental and weather data, will make the model more flexible and useful across various farming situations. This work provides a strong starting point for future AI-based crop health management.