ATTENTION BASED CONDITIONAL GAN FOR SYNTHETIC CROP DATA: SOLVING AGRICULTURE’S DATA AVAILABILITY AND QUALITY CHALLENGES

Abstract

Agriculture is pivotal for the global economy and sustenance. However, it confronts challenges from a burgeoning population, climate shifts, and the imperative for sustainable practices. Artificial intelligence (AI) based solutions offer promise, but the need for substantial-high-quality training data in agriculture is impractical. Existing methods for generating synthetic data face significant challenges regarding precision and reliability, compromising their effectiveness in complex AI-based models for agriculture. To overcome this, we propose an attention-based conditional Generative Adversarial Network enhanced with correlation coefficients from original datasets. Unlike existing methods, our approach effectively replicates the intricacies of real-world agricultural data. Through comprehensive evaluations, we validate its superior performance in producing realistic and relevant synthetic datasets. Incorporating correlation coefficients as a condition and utilizing multi-head attention in the generator, our approach effectively captures the intricate relationships in agricultural data. Leveraging these data enables the training of more precise and accurate models for the agricultural field. Our code is available at: https://github.com/aashikrasool/Coefficient-Based-Data-Generator