High performance practical key terahertz applications.Highpowerterahertzoscillandatorscanberealizedradiationbyutilizsourceingisthethegreatimpedandevicecenforear π,2π and 3π stop-band which means that strong interaction can occur. Compared with the traditional Backward Wave Oscillator(BWO), the Band-edge Oscillator(BO) can achieve high power which can achieve the order of 100 W in W band and 1 W in the THz regime; can realize miniaturized device with Periodic Permanent Magnet(PPM) focusing system; can realize compact structure with Folded Waveguide Slow Wave Structure(FWG SWS) which is short in size and simple in structure; can realize low cost mass production. In this research, a stagger half-period FWG SWS and a novel operation concept for the BO with the folded waveguide are proposed. The BO is composed of Pierce-style double-anode convergent electron gun, periodic permanent magnet focusing system, diamond pill-box window and high efficiency collector. Several novel BOs above 100 GHz have been simulated, fabricated and tested, which achieved output power of 140 W at 100 GHz, 30 W at 120 GHz, and over 1 W at 300 GHz. The dispersion characteristics, design of devices and test results are described and analyzed.

Gaussian software, Density Functional Theory(DFT) and Hartl-Fokker theory(HF) are selected to simulate the absorption spectra of glucose and fructose single molecules in the 0~3 THz band, and the results are compared with that in the reference literatures. The feasibility of using Gaussian software for single molecule absorption spectra calculation is verified, and the intramolecular vibration patterns corresponding to the absorption peaks at different locations are analyzed based on the results obtained by DFT.

The designing process and measurement results of a 220 GHz frequency doubler based on monolithic GaAs Schottky barrier diode are introduced. Multi-anode structure is adopted in this frequency doubler to improve the output power. Eight anodes are mirror symmetrically arranged along the waveguide to form balanced structure. The conversion efficiency and operating bandwidth are improved by adopting various junction capacitances to suppress the difference of scattering parameters among wave ports. The designed frequency doubler is fabricated and measured. Measurement result shows that the conversion efficiency of frequency doubler is above 15% from 204 GHz to 234 GHz and a peak output power of 90.5 mW with conversion efficiency above 20% is demonstrated at 226 GHz. Finally, a 220 GHz frequency doubler with high output power, high conversion efficiency and wide operating bandwidth is realized by adopting multi-anode structure.

Deep Learning(DL) has achieved remarkable results in speech recognition and image object recognition, which has become the main processing method in these fields instead of traditional processing technology. DL is applied to radar target recognition and classification, and has achieved good results as well. Therefore, DL is tried to be applied in radar imaging. Based on the published literatures in recent years and the characteristics of radar imaging, the research progresses of DL in radar imaging are introduced. Its feasibility, sample selection, the generalization, and the evaluation of imaging quality are analyzed. The application prospects of DL in radar imaging are outlooked.

Aiming at the practical application requirements of high precision, lightweight and instant for Specific Emitter Identification(SEI), a lightweight model design for radiation source individual recognition of Automatic Dependent Surveillance-Broadcast(ADS-B) signal is proposed in this paper. Firstly, the signal data is decoded according to the characteristics of the signal data, and the weight of the unbalanced sample is adjusted to improve the sample quality. Then, the small features of different dimensions are obtained by grouping convolution and splicing with the initial features to realize multidimensional complementary feature fusion and parallel synchronization to improve the recognition efficiency. Network model compression and cross-layer connection are implemented by using a Ghost bottleneck structure, which tends to save computing resources while integrating multi-dimensional characteristics. The experimental results show that the proposed algorithm has the advantages of simple structure and low computational load, high recognition rate of 95.2%, and a stable recognition effect in different capacity samples. The proposed design better balances the needs of individual identification accuracy, lightweight and efficiency for SEI.

A background modeling method of infrared dim small target based on improved eigenspace is proposed in order to effectively remove the interference of dynamic background on dim small target signal. Firstly, an improved anisotropic filtering algorithm is employed to filter from the spatial perspective to constrain the differences of each component of the image. Then, a feature matrix is formed from the filtered images in the continuous time domain, and the Principal Component Analysis (PCA) is adopted to perform feature decomposition. Finally, the input image is projected onto the eigenspace for background modeling. As to adapt to the dynamic background, the background model is updated with a certain learning rate in temporal domain. Experimental results show that the proposed algorithm achieves better background estimation effect than the traditional algorithm. The structural similarity SSIM, contrast gain I and background suppression factor BIF are greater than 0.97, 15.46 and 5.25 respectively

Due to lack of a unified standard for performance evaluation of side-lobe blanking, there are few studies on the effect of subarray partition on the side-lobe blanking performance. The performance evaluation parameter of side-lobe blanking—side-lobe blanking ratio is put forward, whose essence is to calculate the proportion of the area satisfying the requirement of the side-lobe blanking in the whole pattern region. The guard channel is constructed by weighted non-phase-coherent accumulation of subarray data directly, which can realize adaptive interference suppression and support the adaptive side-lobe blanking under the interference condition. The simulation results show that the stronger the non-uniformity of subarray partition, the larger the number of the subarray; the smaller the scanning angle, the higher the side-lobe blanking ratio and the better the side-lobe blanking performance of the active phased array antenna.

Microwave technology for brain information detection, which is gradually becoming mature, has been applied to the entity detection and operation. Through the combination of ultrasonic and nuclear magnetic methods, the microwave information detection system has been improved and the efficiency of information detection has also been enhanced. Three kinds of microwave technologies of brain deep information detection and the existing problems are concluded in this paper. It is found that the microwave technology has great potential in brain detection through the analysis on three different microwave detection applications. These applications can effectively provide solutions for pathological examination of brain tissue or other human tissue. By combining with artificial intelligence, microwave technology can also be applied in remote monitoring of the brain or other parts of the body.

The employment of electromagnetic radiation signal is an important detection means in current lightning research, and the accuracy of its positioning results is very important to correctly understand the discharge mechanism of lightning. In this paper, the theoretical model of the Doppler effect on the array and on the localization result of the motion lightning source is built. The results show that the Doppler effect affects at different degrees on the lightning of different types, at different distances and heights. When the speed of the lightning source is different, the Doppler effect is different as well, and the errors cannot be negligible. This work is important for the motion lightning source localization at higher spatiotemporal resolution and the refined study of lightning.

If a side-channel attack can attack the structure and framework of the neural network to recover information such as structure and weight, sensitive information leakage will occur. Therefore, it is important to guard the neural network computing devices against disclosure of sensitive information in the field of side-channel attack. Based on the Jetson Nano platform, a method is designed for the side-channel electromagnetic leakage signal acquisition during the inference of the neural network. The side-channel analysis is conducted by using the deep learning method, and two aspects of security are assessed. Research shows that a good network conversion strategy can improve the classification and recognition accuracy of the network by 5%~12%. In the two evaluation tasks, the classification accuracy of electromagnetic leakage is 97.21% for typical neural network inferences with different structures under the same framework; it reaches 100% for the same kind of network reasoning under different frameworks of neural network. It indicates that the side-channel electromagnetic attack method poses a threat to the privacy of deep learning algorithms in such embedded Graphics Processing Unit(GPU) computing platforms.

Deep learning based face recognition has outperformed traditional methods in many application scenarios. There are two main lines of research to design loss functions for face recognition, i. e., verification and identification. The verification loss functions match the pipeline of open-set face recognition, but it is hard to implement. Therefore, most state-of-the-art deep learning methods for face recognition take the identification loss functions with softmax output units and cross-entropy loss. Nevertheless, identification loss function dose not match the training process with evaluation procedure. A verification loss function is proposed for open-set face recognition to maximize partial area under the Receiver-Operating-Characteristic(ROC) curve, partial Area Under Curve(pAUC). A class-center learning method is also proposed to improve training efficiency, which is critical for the proposed loss function to be comparable to the identification loss in performance. Experimental results on five large scale unconstrained face recognition benchmarks show that the proposed method is highly competitive with state-of-the-art face recognition methods.

In view of the problem of noise interference caused by environmental influence in the process of oil and gas well casing detection, a phased dynamic adaptive noise cancellation method of electromagnetic data of oil and gas well casing is proposed. By comparing the differences among the simulation signal, the detection signal of experimental well and the detection signal of actual well, and combining with the analysis of the noise source in the actual detection process, a detected signal compensation model based on the influence factors of eccentricity jitter temperature is established. Based on the compensation model, a segmented dynamic adaptive noise reduction model of casing electromagnetic testing data is trained. The experimental results show that the method can better reduce the noise information in the logging, which has certain application value.

Speech perception is an important part of unmanned systems. Most of the existing work focuses on the speech perception of a single agent, which is affected by factors such as noise and reverberation, and the performance has an upper limit. Therefore, it is necessary to study multi-agent speech perception, and improve perception performance through multi-agent self-organization and mutual cooperation. A multi-agent ad-hoc speech system is proposed under the assumption that each agent outputs a channel of speech stream. The multi-agent ad-hoc speech system aims to comprehensively utilize all channels to improve perception performance. Taking the speech recognition as an example, a channel selection method that can handle large-scale multi-agent speech recognition is proposed. Specifically, an end-to-end speech recognition stream attention mechanism based on Sparsemax operator is proposed to force the channel weights of noisy channels to zero, and make the stream attention bear the function of channel selection. Nevertheless, Sparsemax would punish the weights of many channels to zero harshly. Therefore, Scaling Sparsemax is proposed, which punishes the channels mildly by setting the weights of strong noise channels to zero only. At the same time, a multilayer stream attention structure is proposed to effectively reduce computational complexity. Experimental results in an unmanned system environment with up to 30 agents under the conformer speech recognition architecture show that the Word Error Rate(WER) of the proposed Scaling Sparsemax is lower than that of Softmax by over 30% on simulation data sets, and by over 20% on semi-real data sets, in test scenarios with mismatched channel numbers.

To solve the problem of high resource surplus rate in the application of scheduling algorithms for global power resource scheduling, a new scheduling algorithm combining user profiles and association rules is proposed to achieve reasonable allocation of global power resources. By using the dual clustering algorithm, the electricity consumption data of all users in the entire dispatch area is analyzed, and a power user profile is constructed to describe the personalized electricity consumption needs of users. Based on the user profiles, a global resource allocation model is established centered on meeting user needs. The resource scheduling subtasks are summarized within the entire domain, the support and confidence between different subtasks are calculated, and association rules are combined to achieve grouping of subtasks. The resources are divided into different domains according to the subtasks group. A secondary dispatching center is set up in each domain, and then is connected with the primary dispatching center in the whole domain to realize the centralized scheduling of resources in the whole domain. The experimental results show that after the application of the proposed scheduling algorithm, the daily global resource surplus rate of the power testing system has significantly decreased, only maintaining around 12%. This algorithm has good practical value.

Aiming at the problem that the current remote sensing satellite Charge Couple Device (CCD) camera has an increasing width and speed, and the output bandwidth of existing camera simulator is insufficient, an Non-Volatile Memory express(NVMe) based ultrahigh-speed multi-channel remote sensing camera simulator is proposed and implemented. This simulator employs a Field Programmable Gate Array(FPGA) to implement four groups of NVMe SSD host controllers and complete the read and write operations on the Solid State Drive(SSD). The Direct Memory Access(DMA) controller is utilized to read the cache data in DDR4, and the data is output through the optical fiber interface after being encapsulated. Experimental results show that the average write rate of the NVMe host controller can reach 1.7 GBps, and the average read rate can reach 3.2 GBps. The overall storage capacity of the simulator is 8 TB, the external output bandwidth is up to 80 Gbps, and it supports eight-way optical fiber interface outputs. This simulator with strong stability and good scalability, has been successfully applied in a remote sensing satellite CCD camera simulator system, which provides sufficient guarantee for the testing and debugging of device such as data transmission.