Objective To enhance the accuracy of detecting fruit surface defects caused by minor mechanical friction during transportation.Methods A surface defect detection model for fused images of fruits based on improved ResNet18 is proposed. To address the challenge of multi-scale feature extraction, the model incorporates a MELA module designed to capture image details more comprehensively. Additionally, an energy function-based regularization loss function is introduced to collaboratively optimize the training process alongside the cross-entropy loss function. Furthermore, a transfer learning strategy is employed to mitigate overfitting.Results Experimental results demonstrated that the improved model exhibits exceptional performance in fruit surface defect detection tasks, achieving the accuracy, precision, recall, and F₁ score as high as 99.53%, 99.60%, 99.58%, and 99.58%, respectively.Conclusion The surface defect detection model for fused images of fruits based on improved ResNet18 significantly enhances the detection accuracy and is suitable for surface defect detection in fused images of fruits.