Abstract：Restoration of and improvement in ecosystems has led to a growth in hare numbers to the detriment of farmland and woodland where they appear. Understanding forage of the hare is imperative to better manage them but proves difficult because most hares forage fruit forests and mountain farmlands in night. In this paper we present a method to monitor real-time forage of the hares using infrared thermal imaging and the improved YOLOV3 target detection method - the real-time infrared hare detection network - YOLO (IR-YOLO) network. The feature-extraction component in the proposed network compressed the depth of the feature-extract in the YOLOV3 in attempts to improve the accuracy of detecting small targets from low-resolution infrared images by using the shallow-layer convolution method. In order to simplify computation and improve detection efficiency and accuracy of small targets, the detection component in the network was based on the CenterNet structure. Considering that hares are more active in autumn, real-time hare forage in the infrared images taken from wheat and rape fields were used to validate the proposed method. The datasets include 6 000 infrared images taken at different scales and environments, and the ratio of the datasets used for training to the sets used for validating was 5:1. The results show that the proposed method identified 75% of hares based on the infrared thermal image videos. The average processing speed of the proposed method is 51 frames/s, the detection rate is 15 percentage points higher than that of YOLOV3, Compared with the Faster-RCNN and RFCN-RESNET101, the proposed method increased processing speed by 45 and 20 percentage points respectively. In summary, the proposed method can quickly and accurately detect the forage of hares in complex environment in night, and it can be used to monitor night-movement of other animals appearing in other ecosystems.