Attention mechanisms have recently shown strong potential in improving the performance of convolutional neural networks (CNNs). This letter proposes a fully convolutional attention block (FCAB) that can be combined with a CNN to refine important features and suppress unnecessary ones in synthetic aperture radar (SAR) images. The FCAB consists of a channel attention module and a spatial attention module. For the channel attention module, we use average-pooling and max-pooling to learn complementary features, and apply group convolution to aggregate the information of the two types of channels. Global average-pooling is then used to encode the channel-wise importance. For the spatial attention module, the average-pooling and max-pooling along the channel axis are used to generate two spatial feature maps, and then two very lightweight convolutional layers are used to encode the spatial weight map. Experimental results on SAR images demonstrate that our FCAB can focus on important channels and object regions. It uses relatively few parameters and is computationally efficient, while bringing about significant performance gain for SAR recognition.
In this paper, we consider the problem of unmanned aerial vehicle (UAV) path planning. The traditional path planning algorithm has the problems of low efficiency and poor adaptability, so this paper uses the reinforcement learning algorithm to complete the path planning. The classic proximal policy optimization (PPO) algorithm has problems that the samples with large rewards in the experience replay buffer will seriously affect training, this situation causes the agent’s exploration performance degradation and the algorithm has poor convergence in some path planning tasks. To solve these problems, this paper proposes a frequency decomposition-PPO algorithm (FD-PPO) based on the frequency decomposition and designs a heuristic reward function to solve the UAV path planning problem. The FD-PPO algorithm decomposes rewards into multi-dimensional frequency rewards, then calculate the frequency return to efficiently guide UAV to complete the path planning task. The simulation results show that the FD-PPO algorithm proposed in this paper can adapt to the complex environment, and has outstanding stability under the continuous state space and continuous action space. At the same time, the FD-PPO algorithm has better performance in path planning than the PPO algorithm.
23 November 2020 in International Journal of Machine Learning and Cybernetics
Ballistic missile defense systems require accurate target recognition technology. Effective feature extraction is crucial for this purpose. The deep convolutional neural network (CNN) has proven to be an effective method for recognizing high-resolution range profiles (HRRPs) of ballistic targets. It excels in perceiving local features and extracting robust features. However, the standard CNN's fully connected manner results in high computational complexity, which is unsuitable for deployment in real-time missile defense systems with stringent performance requirements. To address the issue of computational complexity in HRRP recognition based on the standard one-dimensional CNN (1DCNN), we propose a lightweight network called group-fusion 1DCNN with layer-wise auxiliary classifiers (GFAC-1DCNN). GFAC-1DCNN employs group convolution (G-Conv) instead of standard convolution to effectively reduce model complexity. Simply using G-Conv, however, may decrease model recognition accuracy due to the lack of information flow between feature maps generated by each G-Conv. To overcome this limitation, we introduce a linear fusion layer to combine the output features of G-Convs, thereby improving recognition accuracy. Additionally, besides the main classifier at the deepest layer, we construct layer-wise auxiliary classifiers for different hierarchical features. The results from all classifiers are then fused for comprehensive target recognition. Extensive experiments demonstrate that GFAC-1DCNN with such simple and effective techniques achieves higher overall testing accuracy than state-of-the-art ballistic target HRRP recognition models, while significantly reducing model complexity. It also exhibits a higher recall rate for warhead recognition compared to other methods. Based on these compelling results, we believe this work is valuable in reducing workload and enhancing missile interception rates in missile defense systems.
3 February 2015 in Pattern Analysis and Applications