Research Of Few-Shot Food Recognition

Food recognition has received a significant amount of attention in various fields,such as computer vision,data mining and multimedia communities motivated by many applications in automated food monitoring and dietary management,food trend and popularity analysis smart home and food safety.However,most existing work focuses on food recognition with large amounts of labelled samples,thus fail to robustly recognize food categories with few samples,under this condition,few-shot food recognition is an urgent problem.The main works of the paper are as follows:1.To explore additional ingredient information for few-shot food recognition,a Multi-View Few-Shot Learning(MVFSL)framework is proposed to exploit rich food ingredients for few-shot food recognition..We conduct the comprehensive experimental evaluation on various food benchmarks and experimental results verify the effectiveness of MVFSL,also the experimental results again demonstrate the advantage in exploiting ingredient information.2.To study the impact of more fine-grained differentiation on few-shot food recognition,we use the triplet network to learn the inter-class and intra-class information,however the liner metric function is not discriminative enough for measuring similarities of food images.To address this problem,we use the learnable relation network as non-linear metric and propose a triplet network with relation network to solve the above two disadvantages of the few-shot learning and triplet network.In addition,we proposed an on-line mining rule for triplet samples,which makes the model stable in the training stage.The experimental results verify the effectiveness of the proposed model and sample rule.