Detecting Carcass Defects of Native Chickens Using Convolutional Neural Networks
Journal
2023 ASABE Annual International Meeting
ISBN
9781713885887
Date Issued
2023-01-01
Author(s)
Lai, Pin Cheng
Abstract
Chicken is among the most common meats consumed with high economic value worldwide. To provide high-quality chicken meat to customers, it is necessary to identify carcass defections and excluded the defected carcasses in the chicken slaughter process. In Taiwan, native country chickens fit to the traditional cooking approach and take account for a big portion in meat-type chicken market. Different from the broiler, Taiwan native chickens include breeds with various sizes and external appearances (e.g., blue or black shank and skin). These varieties are usually processed in the same slaughter line. Due to these differences, commercial automatic carcass defect detection systems used for broilers are suboptimal for Taiwan native chickens. Conventional method to pick out the defected carcasses of Taiwan native chickens is through manual identification. However, manual approach is time-consuming and inefficient. Thus, this study proposes to detect the carcass defects of Taiwan native chickens using convolutional neural networks. In this study, a camera system was developed and installed on a slaughter process line. A two-stage detection approach was used to detect defected carcasses automatically. In the first stage, a YOLOv7 model was used to detect four body parts of chicken carcasses, namely the wings, breast, back, and legs. Subsequently, in the second stage, a Resnetv2_x1_bitm model was used to determine whether the body parts were defected. The trained YOLOv7 model achieved a mean average precision of 98.9% on identifying the carcass body parts. The Resnetv2_x1_bitm model achieved an accuracy of 94.2% on classifying defected body parts. The developed system is expected to enhance the efficiency and effectiveness in detecting defect parts of Taiwan native chickens.
Subjects
chicken carcass | Convolutional neural networks | deep learning | defects classification
Type
conference paper