Research Article
Attention Aware Deep Learning Object Detection and Simulation
@INPROCEEDINGS{10.1007/978-3-030-72795-6_1, author={Jiping Xiong and Lingyun Zhu and Lingfeng Ye and Jinhong Li}, title={Attention Aware Deep Learning Object Detection and Simulation}, proceedings={Simulation Tools and Techniques. 12th EAI International Conference, SIMUtools 2020, Guiyang, China, August 28-29, 2020, Proceedings, Part II}, proceedings_a={SIMUTOOLS PART 2}, year={2021}, month={4}, keywords={Deep learning Object detection Convolutional neural network Visualization}, doi={10.1007/978-3-030-72795-6_1} }- Jiping Xiong
Lingyun Zhu
Lingfeng Ye
Jinhong Li
Year: 2021
Attention Aware Deep Learning Object Detection and Simulation
SIMUTOOLS PART 2
Springer
DOI: 10.1007/978-3-030-72795-6_1
Abstract
Dish recognition has certain difficulties in specific applications. Because in the actual inspection, the dishes are filled with food, and the food occupy most of the space of the dishes, and only the edges of the dishes can be seen. If you use empty dishes for training, the accuracy will be low due to insufficient feature matching during actual detection. At the same time, due to the wide variety of foods, if we collect all the food during training, the pre-processing workload will be very large. Based on the above ideas, this paper analyzes the model through three visualization methods, improves Faster R-CNN, and proposes a Cross Faster R-CNN model. This model consists of Faster R-CNN and Cross Layer, which can fuse the low-level features and high-level features of dishes. During training, the model can focus the feature extraction on the edges of the dishes, reducing the interference of food on dish recognition. This method improves the detection accuracy without significantly increasing the detection time. The experimental results show that compared with Faster R-CNN, the accuracy and recall of Cross Faster R-CNN have increased to a certain extent, and the detection speed has basically not changed significantly.
