Title:Differential Evolution Integrated Hybrid Deep Learning Model for Object Detection in Pre-made Dishes

Abstract:With the continuous improvement of people's living standards and fast-paced working conditions, pre-made dishes are becoming increasingly popular among families and restaurants due to their advantages of time-saving, convenience, variety, cost-effectiveness, standard quality, etc. Object detection is a key technology for selecting ingredients and evaluating the quality of dishes in the pre-made dishes industry. To date, many object detection approaches have been proposed. However, accurate object detection of pre-made dishes is extremely difficult because of overlapping occlusion of ingredients, similarity of ingredients, and insufficient light in the processing environment. As a result, the recognition scene is relatively complex and thus leads to poor object detection by a single model. To address this issue, this paper proposes a Differential Evolution Integrated Hybrid Deep Learning (DEIHDL) model. The main idea of DEIHDL is three-fold: 1) three YOLO-based and transformer-based base models are developed respectively to increase diversity for detecting objects of pre-made dishes, 2) the three base models are integrated by differential evolution optimized self-adjusting weights, and 3) weighted boxes fusion strategy is employed to score the confidence of the three base models during the integration. As such, DEIHDL possesses the multi-performance originating from the three base models to achieve accurate object detection in complex pre-made dish scenes. Extensive experiments on real datasets demonstrate that the proposed DEIHDL model significantly outperforms the base models in detecting objects of pre-made dishes.