A Machine Vision-Based Broiler Chicken Weight Detection System Using Deep Learning with Integrated Automatic Feed Regulation

Implementation of a Machine Vision-Based Broiler’s Weight Detector with integrated automatic feed regulation for the improvement of poultry farms is the aim of the research, as the conventional way of weighing is time-consuming and is a stress ignition for both the farmers and the chickens. The researchers gathered 1,236 images that will cover broiler chickens’ weight data from day 1 to day 37. The data set was expanded for data augmentation producing 12,360 photos. The researchers then tested 7 models which are MobileNet, MobileNetV2, InceptionV3, ResNet50, VGG19-based, VGG16-based, and Inception ResNetV2, using metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R2) scores. InceptionResNetV2 model provided the best result with R2 score of 1.00 and MAE of 10.08. While ResNet50 model showed the poorest performance with MAE of 108.76 and R2 score of 0.89. After deciding the best model, it is implemented on the device with a user-friendly graphic user interface. To evaluate the performance of the device, 15 trials were performed on different stages of chickens’ growth. The device resulted with an average overall error of 2.505%, average absolute error of 2.668 grams in weight prediction, and an error of 2.005 grams in feed dispensing.