基于SC注意力机制和集成学习的冰箱食材识别

近年来，随着智能冰箱技术的不断发展，对冰箱果蔬食材进行精准的类别识别，进而对食材进行保鲜控制，得到了研究者越来越多的关注。目标检测技术依靠深度学习相关技术的发展，也渐渐应用于食材盘点的方法。通过对冰箱果蔬食材特性进行分析，提出了一种基于注意力机制和集成学习思想的YOLOv5和EfficientDet融合的方法。首先对冰箱食材数据集进行了伪彩色图像处理，将SE模块和CBAM模块整合提出了新的SC模块，并引入到YOLOv5s网络中，组成SC-YOLOv5s网络结构；然后将SC-YOLOv5s网络结构与EfficientDetd0网络进行异质集成；最后用集成后的整体网络对尺度有差异但外貌相似的食材进行识别。实验结果表明当IOU阈值为0.5时，在60类果蔬食材测试集上，改进后集成模型的平均最大精确度(mAP)从SC-YOLOv5s的95.88%和EfficientDetd0的83.22%提高到了97.36%，明显提升了对果蔬类食材的检测效果。

Recognition of Refrigerator Ingredients Based on SC Attention Mechanism and Ensemble Learning

In recent years, with the continuous development of smart refrigerator technology, accurate category identification of refrigerator fruit and vegetable ingredients and fresh-keeping control of ingredients have attracted more and more attention from researchers. With the development of artificial intelligence technology, target detection technology has become a method of food material inventory. By analyzing the characteristics of fruits and vegetables in the refrigerator, a fusion method of YOLOv5 and EfficientDet based on attention mechanism and integrated learning is proposed. Firstly, the pseudo color image processing is carried out on the refrigerator food material dataset, the SE module and CBAM module are integrated, and a new SC module is proposed, which is introduced into the YOLOv5s to form the SC-YOLOv5s structure. Then, the SC-YOLOv5s structure is heterogeneously integrated with EfficientDetd0. Finally, the ensemble network is used to identify the ingredients with different scales but similar appearance. The experimental results show that when the IoU threshold is 0.5, on the test set of 60 kinds of fruits and vegetables, the mean average precision (mAP) of the improved ensemble model is improved from 95.88% of SC-YOLOv5s and 83.22% of EfficientDetd0 to 97.36%, which improves the detection effect of fruits and vegetables.