A Comparative Study of Deep Convolutional Neural Network Architectures to Identify Full Bee Body in Images

This paper presents a comprehensive comparison of deep convolutional neural network (CNN) architectures and TensorFlow for full bee body identification in images with the purpose of using detected individual bees for future health analysis. The main focus is on five popular architectures: Faster R-CNN ResNet152, Faster R-CNN Inception ResNet, SSD ResNet50, SSD Mobilenet, and CenterNet Resnet50. The results show that SSD Mobilenet model offers the highest recall value of 80.5 percent but with the lowest precision among tested models of 95 percent. In contrast, Faster R-CNN ResNet152 is the second-best option with a recall value of 67.6 percent, the fastest image processing time, and a precision of 99.6 percent.