A two-stage approach to agricultural products authentication by mining subtle local features

Abstract

Counterfeit agricultural products recognition poses a significant challenge due to the high visual similarity between genuine and fake items. Existing methods often struggle to capture the subtle details necessary for reliable differentiation. This paper presents Focus on Detail (FoD), a novel approach that emphasizes the automatic discovery of trustworthy 'authenticity cues' on the products. By employing custom-designed loss functions and a semi-supervised training strategy, FoD learns to suppress distracting background regions and focus exclusively on the most critical local features. Experimental results demonstrate that FoD achieves superior performance on standard benchmark datasets, establishing a new state-of-the-art in both accuracy and speed.