Core body temperature (CBT) measurement of laying hens is very complex under cage breeding conditions. Meanwhile, traditional measurement methods also require handing the hens, can be stressful. Infrared thermography is an alternative means for assessing hens core temperature. A method was proposed for estimating the CBT of laying hens using infrared thermography and deep learning. A total of 10994 infrared thermal images and corresponding CBT were collected through 172 hens. The hens facial were selected as region of interest (ROI). The YOLO v8s object detection algorithm was employed to automatically identify the ROI within the images. Additionally, the modified Res2Net50 network was used for regression training between ROI images and CBT values. Then the above two algorithms were combined to directly estimate the CBT of laying hens using infrared thermal images. Comparative experiments were conducted with four object detection algorithms (YOLO v4s, YOLO v5s, YOLO v7, YOLOX-s), and the results indicated that YOLO v8s achieved superior precision (99.38%), mAP(99.9%), and recall(99.87%), compared with the other algorithms. Furthermore, seven algorithms (MobileNetV3, GhostNet, ShuffleNetV2, RegNet, ConvNeXt, Res2Net, MobileVIT) were compared with the modified Res2Net, and the results demonstrated that the modified Res2Net exhibited a higher coefficient of determination (R2) of 0.97364 and adjusted coefficient of determination (R2adj) of 0.97352 on the test images, surpassing the other algorithms. Finally, CBT estimation experiments were conducted by using the YOLO v8s-Res2Net50 algorithm. Nine layers were randomly selected, and their infrared thermal images were input into the algorithm network. The results showed that the ROI could be fully identified, and the mean absolute error (MAE) of estimating CBT was 0.153℃. Thus the proposed deep learning model for CBT estimation can offer an effective automated detection method for assessing CBT in laying hens.