In this paper, a deep neural network with concatenated structure is created for the recognition of flight targets. Compared with the traditional recognition method, the deep network model automatically gets deeper structure information that is more useful to the classification, and better performance of target recognition is also obtained when using high resolution range profile (HRRP) for radar automatic target recognition (RATR). First, the framework is expanded and cascaded by multiple shallow neural sub-networks. Then a secondary-label coding method is proposed to solve the target-aspect angle sensitivity problem. The samples are divided into sub-classes based on aspect angle, each of which is assigned a separate encoding bit in category label. Finally, the recognition results of multiple samples are fused by multi-evidence fusion strategy for the improvement of recognition rate. Furthermore, the effectiveness of the proposed algorithm is demonstrated on the measured and simulated data.

Soil moisture (SM) retrieval using wireless signals has become a research focus with the development of sensor devices in internet of things (IOT). Studies applying ground-penetrating radar (GPR) have improved the accuracy of SM retrieval, however the field-scaled data are hardly satisfactory mainly due to the frequency response of the antenna, and it’s not cost-effective for farmers to monitor the soil conditions. In this paper, we compare two fuzzy logic systems (FLS) -type-1 fuzzy logic system (T1FLS) and adaptive network-based fuzzy inference system (ANFIS) to extract fuzzy parameters of soil. Moreover, two machine learning (ML) algorithms - random forest (RF) and artificial neural network (ANN) with principal component analysis (PCA) are applied in the SM classifications. 9 types of UWB soil echoes of different texture and volume water content (VWC) are collected and investigated using our approaches. Final analysis shows that ANFIS with RF provides the best VWC correct recognition rate (CRR) compared to other algorithms.

Jamming will seriously affect the detection ability of radar, so it is essential to suppress the jamming of radar echoes. Interrupted-sampling repeater jamming (ISRJ) based on a digital-radio-frequency-memory (DRFM) device can generate false targets at the victim radar by the interception and repeating of the radar transmission signal, which is highly correlated with the true target signal. ISRJ can achieve main lobe jamming and has both deception and oppressive jamming effects, so it is difficult for the existing methods to suppress this jamming effectively. In this paper, we propose a deep-learning-based anti-jamming network, named MSMD-net (Multi-stage Multi-domain joint anti-jamming depth network), for suppressing ISRJ main lobe jamming in the radar echo. In the first stage of MSMD-net, considering that the target signal is difficult to detect under a high jamming-to-signal ratio (JSR), we propose a preprocessing method of limiting filtering on the time-frequency domain to reduce the JSR using the auxiliary knowledge of radar. In the second stage, taking advantage of the discontinuity of the jamming in the time domain, we propose a UT-net network that combines the U-net structure and the transformer module. The UT-net performs target feature extraction and signal reconstruction in the signal time-frequency domain and preliminarily realizes the suppression of the jamming component. In the third stage, combined with phase information, a one-dimensional complex residual convolution U-net network (ResCU-net) is constructed in the time domain to realize jamming filtering and signal recovery further. The experimental results show that MSMD-net can obtain the best jamming suppression effect under different transmitted signals, different jamming modes, and different jamming parameters.

Existing deep learning algorithms for synthetic aperture radar (SAR) image recognition are performed with offline data. These methods must use all data to retrain the entire model when new data are added. However, facing the real application environment with growing data, retraining consumes much time and memory space. Class-Incremental Learning (CIL) addresses this problem that deep learning faces in streaming data. The goal of CIL is to enable the model to continuously learn new classes without using all data to retrain the model while maintaining the ability to recognize previous classes. Most of the CIL methods adopt a replay strategy to realize it. However, the number of retained samples is too small to carry enough information. The replay strategy is still trapped by forgetting previous knowledge. For this reason, we propose a CIL method for SAR images based on self-sustainment guidance representation. The method uses the vision transformer (ViT) structure as the basic framework. We add a dynamic query navigation module to enhance the model’s ability to learn the new classes. This module stores special information about classes and uses it to guide the direction of feature extraction in subsequent model learning. In addition, the method also comprises a structural extension module to defend the forgetting of old classes when the model learns new knowledge. It is constructed to maintain the representation of the model in previous classes. The model will learn under the coordinated guidance of old and new information. Experiments on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset show that our method performs well with remarkable advantages in CIL tasks. This method has a better accuracy rate and performance dropping rate than state-of-the-art methods under the same setting and maintains the ability of incremental learning with fewer replay samples. Additionally, experiments on a popular image dataset (CIFAR100) also demonstrate the scalability of our approach.

Two main difficulties for the accurate direction of arrival (DOA) estimation in a distributed sensor system are grating lobes and system synchronization. Different from the traditional phased array system based on single carrier frequency, this paper designs a novel architecture for distributed sensor system using dual carrier frequency measurements. Both two carrier frequencies are combined and transmitted simultaneously in the proposed architecture. Instead of single carrier frequency $f_1$ in the traditional method, the frequency separation $f_2-f_1$ is used in the proposed algorithm to determine a maximum unambiguous DOA estimate. Since $f_2-f_1$ is much less than $f_1$ and the choice of $f_2-f_1$ is more flexible, the proposed method can provide an unambiguous DOA estimate in the whole field of view (FOV) for arbitrary element spacing and a looser requirement on system synchronization than the traditional method based on single carrier frequency. For a case with $f_1=3$ Ghz, $f_2-f_1=150$ Mhz and 10 times element spacing, there is no grating lobe for the proposed method in the whole FOV $[-\pi /2,\pi /2]$, whereas the traditional method has many grating lobes and the FOV of the traditional method without grating lobes is only ${5.73}^{\circ }$. To obtain the same combining efficiency in this case, the requirement on the positioning error of the proposed method is 0.4 m and the location error of the traditional method must be smaller than 0.02 m. Hence, the proposed method can greatly reduce hardware requirements. The performances of the proposed architecture for the cases with perfect and imperfect system synchronizations are evaluated in terms of the average beampattern and combining efficiency, respectively. Some characteristics of the derived beampattern and combining efficiency are also presented in the paper.

Recent work has shown success in incorporating pre-trained models like BERT to improve NLP systems. However, existing pre-trained models lack of causal knowledge which prevents today's NLP systems from thinking like humans. In this paper, we investigate the problem of injecting causal knowledge into pre-trained models. There are two fundamental problems: 1) how to collect various granularities of causal pairs from unstructured texts; 2) how to effectively inject causal knowledge into pre-trained models. To address these issues, we extend the idea of CausalBERT from previous studies, and conduct experiments on various datasets to evaluate its effectiveness. In addition, we adopt a regularization-based method to preserve the already learned knowledge with an extra regularization term while injecting causal knowledge. Extensive experiments on 7 datasets, including four causal pair classification tasks, two causal QA tasks and a causal inference task, demonstrate that CausalBERT captures rich causal knowledge and outperforms all pre-trained models-based state-of-the-art methods, achieving a new causal inference benchmark.