Body temperature serves as a crucial indicator of dairy cow health. Traditional rectal temperature (RT) measurement often induces stress responses in animals. Body temperature detection based on infrared thermography (IRT) offers non-invasive and timely advantages, contributing to welfare-oriented farming practices. However, automated detection and temperature extraction from critical cow regions are susceptible to complex illumination, black-and-white fur texture interference, and region of interest (ROI) deformation, resulting in low detection accuracy and poor robustness. To address this, this paper proposes the EDC-YOLO-World-DB framework to enhance detection and temperature extraction performance under complex illumination conditions. First, URetinex-Net and CLAHE methods are employed to enhance low light and overexposed images, respectively, improving structural information and boundary contour clarity. Subsequently, spatial relationship constraints between LU and AA are established using five-class text priors—lower udder (LU), around the anus (AA), rear udder, hind legs, and hind quarters—to strengthen the spatial localisation capability of the model for ROIs. Subsequently, a Dual Bidirectional Feature Pyramid Network architecture incorporating EfficientDynamicConv was introduced at the neck of the model to achieve dynamic weight allocation across modalities, levels, and scales. Task Alignment Metric, Gaussian soft-constrained centroid sampling, and combined IoU (C_IoU + G_IoU) loss were introduced to enhance sample matching quality and regression stability. Results demonstrate detection confidence improvements by 0.08 and 0.02 in low light and overexposed conditions, respectively; compared to two-text input, five-text input increases P, R, and mAP50 by 3.61%, 3.81%, and 1.67%, respectively; Comprehensive improvements yielded P = 88.65%, R = 85.77%, and mAP50 = 89.33%—further surpassing the baseline by 2.79%, 3.01%, and 1.92%, respectively. Temperature extraction experiments demonstrated significantly reduced errors for $T_{Max}$, $T_{Min}$, and $T_{avg}$. Specifically, for the mean error of LU, $T_{Max}$, $T_{Min}$, and $T_{avg}$ were reduced by 66.6%, 33.5%, and 4.27%, respectively; for AA, $T_{Max}$, $T_{Min}$, and $T_{avg}$ were reduced by 66.6%, 25.4%, and 11.3%, respectively. This study achieves robust detection of LU and AA alongside precise temperature extraction under complex lighting and deformation conditions, providing a viable solution for non-contact, low-interference dairy cow health monitoring.

Dairy cow body temperature is a key physiological indicator that reflects metabolic level, immune status, and environmental stress responses, and it has been widely used for early disease recognition. Infrared thermography (IRT), as a non-contact imaging technique capable of remotely acquiring the surface radiation temperature distribution of animals, is regarded as a powerful alternative to traditional temperature measurement methods. Under practical cowshed conditions, IRT images of dairy cows are easily affected by complex background interference and generally suffer from low resolution, poor contrast, indistinct boundaries, weak structural perception, and insufficient texture information, which lead to significant degradation in target detection and temperature extraction performance. To address these issues, a lightweight detection model named YOLOv11n-CGSD is proposed for dairy cow IRT images, aiming to improve the accuracy and robustness of region of interest (ROI) detection and body temperature extraction under complex background conditions. At the architectural level, a C3Ghost lightweight module based on the Ghost concept is first constructed to reduce redundant feature extraction while lowering computational cost and enhancing the network capability for preserving fine-grained features during feature propagation. Subsequently, a space-to-depth convolution module is introduced to perform spatial rearrangement of feature maps and achieve channel compression via non-strided convolution, thereby improving the sensitivity of the model to local temperature variations and structural details. Finally, a dynamic sampling mechanism is embedded in the neck of the network, where the upsampling and scale alignment processes are adaptively driven by feature content, enhancing the model response to boundary temperature changes and weak-texture regions. Experimental results indicate that the YOLOv11n-CGSD model can effectively shift attention from irrelevant background regions to ROI contour boundaries and increase attention coverage within the ROI. Under complex IRT conditions, the model achieves P, R, and mAP50 values of 89.11%, 86.80%, and 91.94%, which represent improvements of 3.11%, 5.14%, and 4.08%, respectively, compared with the baseline model. Using T_max as the temperature extraction parameter, the maximum error (Max. Error) and mean error (MAE. Error) in the lower udder region are reduced by 33.3% and 25.7%, respectively, while in the around the anus region, the Max. Error and MAE. Error are reduced by 87.5% and 95.0%, respectively. These findings demonstrate that, under complex backgrounds and low-quality IRT imaging conditions, the proposed model achieves lightweight and high-performance detection for both lower udder (LU) and around the anus (AA) regions and provides a methodological reference and technical support for non-contact body temperature measurement of dairy cows in practical cowshed production environments.