The lightweight model for salient object detection in optical remote sensing images (SOD-RSI) is a recent emerging topic. Due to the complexity of the task, recently published works have achieved effective model compression but have not yet achieved the desired detection speed. To truly release the detection speed of lightweight models while ensuring a favorable accuracy-efficiency tradeoff, we propose a new speed-oriented lightweight SOD-RSI network (SOLNet), which has significant advantages in detection speed. Specifically, we design a lightweight group attention (LGA) module to deconstruct–interact–recombine channel features and an enhanced dynamic encoding (EDE) module for dynamically capturing spatial information. On this basis, the dynamically enhanced aggregation module (DEAM) is further proposed, which mines the intrinsic correlation of feature information by decoding high-level feature maps, eliminating the need to pay additional attention to other scales. SOLNet completes lightweight and efficient decoding through simple cascade aggregation operations. Notably, we also propose an evaluation strategy that takes both speed and accuracy into account, extending a novel lightweight gain (Lg) metric for SOD-RSI. This not only effectively reveals the under-gain issue of lightweight models but also provides theoretical support for the evaluation of subsequent lightweight works. Experimental results on the challenging EORSSD and ORSSD datasets show that SOLNet achieves significant speed improvements and is the state-of-the-art (SOTA) lightweight SOD-RSI method. The code is available at https://github.com/SpiritAshes/SOLNet .

In view of the poor performance of traditional feature point detection methods in low-texture situations, we design a new self-supervised feature extraction network that can be applied to the visual odometer (VO) front-end feature extraction module based on the deep learning method. First, the network uses the feature pyramid structure to perform multi-scale feature fusion to obtain a feature map containing multi-scale information. Then, the feature map is passed through the position attention module and the channel attention module to obtain the feature dependency relationship of the spatial dimension and the channel dimension, respectively, and the weighted spatial feature map and the channel feature map are added element by element to enhance the feature representation. Finally, the weighted feature maps are trained for detectors and descriptors respectively. In addition, in order to improve the prediction accuracy of feature point locations and speed up the network convergence, we add a confidence loss term and a tolerance loss term to the loss functions of the detector and descriptor, respectively. The experiments show that our network achieves satisfactory performance under the Hpatches dataset and KITTI dataset, indicating the reliability of the network.