AMCD: A Multi-Domain Agricultural Crop and Flower Image Dataset for Deep Learning-Based Classification

The Agricultural Multidisciplinary Collection Dataset (AMCD) contains 5,405 JPG images of agricultural crops and flowers collected in Bangladesh. The images are organized into four domains: fruits, vegetables, flowers, and crops/grains. These domains contain 77 subclasses representing commonly observed agricultural and floricultural specimens. Images were captured manually using smartphone cameras at farms, marketplaces, and gardens in Savar, Dhamrai, and Manikganj in the Dhaka Division of Bangladesh between 27 January 2025 and 20 May 2025. The data include natural outdoor lighting, variable backgrounds, different viewpoints, single-object scenes, and multi-object scenes. After collection, images were cleaned, resized to 512 × 512 pixels, color balanced, contrast enhanced, and edge sharpened. Conservative non-synthetic augmentation was applied to underrepresented subclasses using horizontal flipping, small rotations, and brightness adjustment. The dataset can be reused for agricultural image classification, transfer learning, model benchmarking, lightweight mobile model development, domain adaptation, and healthy-specimen reference data in plant image analysis.