Identifying food items on plates is difficult. There is a wide variety of foods. Each food item can be complex. Because different backgrounds on the plates also create difficulties. Additionally, food items can be occluded by other objects. These factors make detection more complex. In this paper, a hybrid model is proposed to overcome such problems. The hybridization of two state-of-the-art deep learning models: Mask R-CNN and YOLOv9 are used for food detection on plates. These models are evaluated against a broad set of plate images based on metrics including mean average precision (mAP), precision, recall and inference speed. The result shows the significant improvement that the hybrid model performs better, with a score of 91.57% mAP and a 90.17% F1 score. The result is better than other state of art which successfully balancing speed and accuracy in detection and segmentation in complex scenarios. After analyzing the proposed model the potential suggestions for future directions are suggested, such as handling occlusions and increasing robustness for small objects.