<p>Rock mechanics and geomechanical studies provide crucial information for production, life, and property safety. To enable intelligent identification of localized enhanced infrared radiation (IR) during rock failure, the IR changes caused by rock failure are observed using an indoor experimental system with advanced shadow treatment techniques. Marble and granite specimens are selected for IR observation experiments under uniaxial compression conditions. A deep learning intelligent model (DLIM) is employed to detect the localized enhanced IR during rock failure and is further applied for remote non-contact identification of landslides. The results show that the transition from global deformation to localized fracture corresponds to the specific features of enhanced IR, especially during critical events leading to rock instability. Using a U-Net network model integrated with the visual geometry group (VGG) architecture and the Adam algorithm, high-precision intelligent detection of localized enhanced IR is achieved, with a mean pixel accuracy (mPA) of 94.47% and a mean intersection-over-union (mIoU) value of 89.21%. Based on the locally enhanced infrared DLIM, an intelligent landslide identification and early warning system is developed, enabling the remote, non-contact monitoring of slope stability. These achievements provide theoretical basis for the intelligent identification of localized enhanced IR and the related application in landslide monitoring and warning systems.</p>