Meal target detection based on improved YoloV3 Algorithm

In recent years, in order to better solve the nursing problems of the existing aging population and the disabled, meal-assistance robots have gradually become a hot spot in the field of service robots. In this paper, an improved Yolov3 model is proposed for the problem of meal target detection during the meal-taking process of the meal-assistance robot. And based on the improved Retinex algorithm, this paper preprocesses the dataset images. By adding the AQ module to the feature extraction network and the MQ prediction layer, this paper designs an improved Yolov3 model. Based on the enhanced meal image dataset, the model is trained, validated and tested. The test results show that the model improves detection accuracy by 4.15% and detection speed by 16.1 % compared with the unimproved model on the same test task. The experimental results show that the improved model further improves the accuracy of the original model while improving the original detection speed, which achieves the expected goal, and meets the meal target detection task requirements of the meal-assistance robot.