Effective inventory management is very much essential to restaurants’ operational performance since service quality and profitability are directly impacted by the need to maintain ideal stock levels of different food items. Traditional methods are usually costly, time-consuming, and susceptible to human error. The emergence of sophisticated machine learning methods, which include the YOLO (You Only Look Once), the commonly used model for object detection, presents a revolutionary way to improve and automate inventory management systems. In this paper, YOLOV8 and YOLOV9 will be used for detection and YOLOV8 for segmentation of the different food items in the restaurants. By integrating these models with an automated image capture and analysis system, restaurant inventory management systems can check stock levels in real-time that guarantee timely replenishment, minimize food waste, and maximize order placements. The mAP (mean average precision) values obtained in each case will be compared to see how well the model performs with each case and also with increasing or decreasing the epochs. The food along with its name and boundary accuracy has been detected. This study evaluates the effectiveness of YOLOV8 and YOLOV9 with respect to detection accuracy, model size, and inference time using two different custom food datasets. The results from our study show that although YOLOV9 is marginally more accurate than YOLOV8, YOLOV8 has a lower model size and a faster pace of inference and how crucial it is of implementing cutting-edge AI technology to optimize processes, boost productivity, and eventually enhance dining experiences through improved inventory control.