Improving passion fruit yield estimation with multi-scale feature fusion and density-attention mechanisms in smart agriculture

The present study proposes a passion fruit yield prediction model based on camera features and semantic segmentation to address challenges in fruit recognition and yield estimation. The model integrates a density-attention mechanism and cross-multi-scale feature fusion, enhancing detection accuracy in complex backgrounds, small fruit instances, and high-density fruit scenarios. Experimental results demonstrate that the proposed model outperforms existing state-of-the-art (SOTA) models across multiple metrics, achieving a precision of 0.88, a recall of 0.83, and an accuracy of 0.85, with an F1-score of 0.85. These results indicate superior robustness and generalization capability, providing novel methodological and theoretical support for fruit detection and yield estimation in smart agriculture. • Proposed a passion fruit yield prediction model using density-attention and multi-scale feature fusion. • Enhanced small fruit detection and dense region recognition in complex agricultural environments. • Integrated camera features with semantic segmentation for precise pixel-to-physical size conversion.