Abstract Aiming to detect novel objects from only a few annotated samples, few-shot object detection (FSOD) has undergone remarkable development. Previous works rarely pay attention to the perspective of gradient propagation to optimize existing methods, therefore failing to make full use of information for novel objects in gradient propagation. We propose a method to solve this problem based on two-stage fine-tuning. A domain adaptation module with multi-constraints is used to promote the spread of gradients, a classification promotion network is used to improve the effect of classification, and a multi-path mask head is added to enrich RoI features. Experiments on PASCAL VOC and COCO datasets show that our model significantly raises the performance compared with previous methods (up to 1–5 $$\%$$ % in average).

Airborne multi-antenna SAR urgently needs the high-precision motion information of each antenna to realize high-precision imaging. Distributed Position and Orientation System (POS) is an important way to solve this problem based on transfer alignment. Because multiple antennas are usually installed on the flexible and discontinuous structure, there are complex and time-varying relative motions between different antennas. This relative motion will lead to the decline of transfer alignment accuracy, and directly affects the imaging accuracy of SAR. To solve this problem, a motion information acquisition algorithm of multi-antenna SAR installed on flexible and discontinuous structure based on distributed POS is given. Firstly, the gyroscopes and accelerometers of the distributed POS are used to calculate the relative attitude between nodes, which is applicable to both continuous and discontinuous structure. Secondly, combining the smooth estimation and the distributed filter, a distributed smoothing algorithm including forward filter and backward smoothing is proposed to calculate the high-precision and smoothed motion data of each sub-node. The forward filter obtains the motion parameters of each sub-node through transfer alignment based on the above relative attitude data. The backward smoothing further improves the motion information accuracy of sub-node by using the measurement data at all time and sharing the data of all nodes. Through the flight imaging test of multi-antenna SAR installed on a flexible and discontinuous platform, it is shown that the proposed algorithm can significantly improve the accuracy of the sub-node motion data, in which the errors of position, velocity and attitude are reduced by 58.9%, 61.9% and 71.2% respectively. Based on this motion data, SAR achieves better imaging focusing effect.

When capturing images in low light condition, due to insufficient lighting, the true color information and texture details of objects are difficult to obtain. Considering that, we propose an end-to-end color channel fusion network (CCFN). Specifically, our proposed method uses partial channel combination inputs to obtain multiple enhancement results. The relevance among RGB channels is maintained by modeling channel interdependencies. Subsequently, a multi-scale feature channel shuffle module (MFCS) is designed to combine image features at different scales, which makes the fusion images hold more rich information. Finally, the output images are generated after detail enhancement. Extensive experiments demonstrate the superiority of our method over several state-of-the-arts in terms of enhancement quality.