Pig-ear detection from the thermal infrared image based on improved YOLOv8n

In the current pig scale breeding process, considering the low accuracy and speed of the infrared thermal camera automatic measurement concerning the pig body surface temperature, this paper proposes an improved algorithm for target detection of the pig ear thermal infrared image based on the YOLOv8n model. The algorithm firstly replaces the standard convolution in the CSPDarknet-53 and neck network with Deformable Convolution v2, so that the convolution kernel can adjust its shape according to the actual situation, thus enhancing the extraction of input features; secondly, the Multi-Head Self-Attention module is integrated into the backbone network, which extends the sensory horizons of the backbone network; finally, the Focal-Efficient Intersection Over Union loss function was introduced into the loss of bounding box regression, which increases the Intersection Over Union loss and gradient of the target and, in turn, improves the accuracy of the bounding box regression. Apart from that, a pig training set, including 3,000 infrared images from 50 different individual pigs, was constructed, trained, and tested. The performance of the proposed algorithm was evaluated by comparing it with the current mainstream target detection algorithms, such as Faster-RCNN, SSD, and YOLO families. The experimental results showed that the improved model achieves 97.0%, 98.1% and 98.5% in terms of Precision, Recall and mean Average Precision, which are 3.3, 0.7 and 4.7 percentage points higher compared to the baseline model. At the same time, the detection speed can reach 131 frames per second, which meets the requirement of real-time detection. The research results show that the improved pig ear detection method based on YOLOv8n proposed in this paper can accurately locate the pig ear in thermal infrared images and provide a reference and basis for the subsequent pig body temperature detection.