Esophagus Detection Using Deep Learning Method

Abstract

The halal food industry has a high demand in halal meat and poultry especially in Muslim countries. In order to slaughter a chicken according to the Islamic Law, it is required to sever the trachea, esophagus and both the carotid arteries and jugular veins to accelerate the chicken's bleeding and death. Syariah Compliance Automated Chicken Processing System (SYCUT) uses the Vision Inspection Technology which is built for the purpose of detecting and classifying whether a chicken is halal or not. The previous work on the system faced a few challenges regarding the image conditions which negatively affected the detection results. This paper discusses the possibility of deep learning approach to combat the challenges and its potential for esophagus detection. The deep learning model used is RetinaN et-MaskRCNN with ResNet50 as the backbone. The evaluation of the trained model yields 92.8% mean average precision (mAP) which performs better than the previous work. The model has a high recall value but a low precision value due to multi-detections.