To resolve the problems of the difficulties in detecting the salient object area under poor lighting conditions in an RGB image and the salient object boundary because of the infrared thermal radiation in a thermal infrared (T) image, a salient detection algorithm based on the complementary fusion of RGB and T image information is proposed. First, image-lighting conditions are established based on the IHS color space to determine the lighting conditions of RGB and T images. An RGB-T image adaptive lighting fusion algorithm is proposed to guide the fusion of RGB-T images according to the lighting conditions of the images, generating a multilevel RGB-T lighting fusion image to improve the ability of salient object area detection. Second, the Gaussian filters with different convolution kernels and standard deviations are used to extract the high-frequency information of the objects’ edges in the RGB and T images, generating different levels of RGB and T high-frequency detailed images. The deep learning network based on encoder-decoder is used to fuse RGB and T high-frequency detail images to generate RGB-T detail fusion images with different levels, which improves the boundary detection ability of salient objects. Finally, a multilevel fusion of RGB-T lighting fusion image and RGB-T detail fusion image is performed according to the image-lighting information, and the algorithm based on learning is used for salient object detection. Experimental results show that the proposed algorithm improves the detection accuracy of the salient area and boundary detections. The proposed algorithm is competitive compared with EGNet, PoolNet, CPDNet, DMRA, and A2dele, which are excellent salient object detection algorithms in F_max value, F_ave value, and average absolute error.