AFFU-Net: Attention feature fusion U-Net with hybrid loss for winter jujube crack detection

•We proposed an attention feature fusion U-Net, named AFFU-Net, to strengthen the feature learning ability and improve the accuracy of crack detection.•A residual refinement module (RRM) was introduced to refine the coarse saliency map and obtain a predicted map with more accurate boundaries.•A novel hybrid loss function was designed to supervise and optimize network training.•We built a quantitative and qualitative evaluation method for winter jujube crack defect assessment.