Generative AI-Driven Synthetic Image Generation for Horticultural Seed Bag Detection and Classification

Abstract

Horticultural seed bags are widely used in agricultural production, storage, transportation, and distribution, making their accurate identification and classification important for effective inventory management and quality-control activities. Conventional image-based approaches often depend on limited collections of real images, which may not adequately represent variations in orientation, lighting, background, packaging appearance, scale, occlusion, and camera viewpoint. This study proposes a Generative AI-driven approach for synthetic image generation and automated detection and classification of horticultural seed bags. The proposed framework employs Generative AI techniques, including GAN-based and diffusionbased models, to learn visual characteristics from available seed-bag images and generate realistic synthetic samples with diverse visual conditions. The generated images are subjected to quality verification based on realism, diversity, visual consistency, and preservation of important seed-bag characteristics. Selected synthetic images are then integrated with the original dataset to create an enhanced image collection for detection and classification. The approach is designed to increase the visual diversity of the available dataset while reducing dependence on extensive manual image collection. The generated outputs are further analyzed according to their usefulness for seed-bag identification and classification. The proposed framework demonstrates the potential of Generative AI for developing flexible and scalable solutions for horticultural image analysis, with applications in agricultural storage, packaging verification, inventory management, and automated quality-control operations.