The visual characteristics of low-altitude drones are less obvious and the scale changes during the detection process. Traditional detection methods are susceptible to interference during detection, and most of those methods cannot work quickly and robustly. To solve this problem, a real-time drone detection algorithm combined with the improved YOLOv3 model and the super resolution method is proposed in this paper. First, frame difference is used to propose the candidate area, and the super-resolution method is used to strengthen the details. Then the dimensional clustering algorithm is used to regenerate the anchors for the model, and the model is slightly adjusted. Finally, we use the improved YOLOv3 to scan both the whole frame and the processed candidate area so as to detect the drones. The frame relationship is also used to implement tracking of drones by real-time detection. With GPU (GTX 1070Ti) acceleration, the method works at a speed of about 20FPS and has an accuracy rate of 96.8% and a recall rate of 95.6%. The results prove that the method can detect drones in different complex backgrounds with a considerable effective detection distance. Compared with the traditional method or normal machine learning method, our method is of a certain theoretical and practical value.
Because existing ultrasonic transducers mostly use PZT and ZnO materials as piezoelectric thin films, while the PZT contains lead and ZnO has the problem of contaminating CMOS manufacturing, a piezoelectric ultrasonic micromechanical transducer with circular bi-laminate bending vibration which uses the aluminium nitride as the piezoelectric layer is designed. The working principle of the transducer is analyzed, the finite element model is established, and the finite element simulation is carried out for the size parameters of the transducer. It is found that the resonant frequency of the transducer is proportional to the thickness of each layer and inversely proportional to the square of the radius of the transducer; when the radius of the upper electrode is about 65% of the radius of the transducer, the resonant amplitude of the transducer is the largest; when the thickness ratio of the silicon and the aluminum nitride of the piezoelectric layer is about 0.6, the resonant amplitude is also the largest. The optimized transducer is simulated and compared with the original model. The results show that the working frequency in air is 9.21MHz, the electromechanical coupling coefficient increases from 21.44% to 27.16% in air and from 3.55% to 11.93% in water. These conclusions provide basic data for the research on the medical imaging probe.
Aiming at the low prediction accuracy of traditional network security situation prediction technology, a network security situation adaptive prediction model (NAP) is proposed. First, it extracts alarm elements and calculate network security situation time sequences based on the entropy correlation method. Then, the sequences are taken as the input of the sliding adaptive cubic exponential smoothing method with initial security situation predicted value sequences generated. Finally, the time-varying weighted Markov chain is used to predict the error value based on the error state and the initial predicted values are modified. Experimental results show that the NAP has a better prediction accuracy than other existing models.
Due to the moving platforms, the clutters in distributed airborne MIMO radar are non-Gaussian and non-homogeneous, which leads to having no independent and identically distributed training data to estimate the clutter covariance matrix. To solve the problem, we propose that the covariance of the clutter should be modeled as an inverse complex Wishart distribution whose average value is a Hadamard product of the covariance matrix taper (CMT) and the clutter Doppler spectrum component. Based on this clutter model, a novel detector combing the Bayesian approach and the generalized likelihood ratio test(GLRT) is proposed. Numerical simulation results show that the proposed detector has a better detection performance compared with two current commonly used non-Bayesian detectors.
An improved state prediction algorithm for edge layer nodes is proposed to solve the problem of the existing state prediction algorithm for edge layer nodes based on Hidden Markov, such as the subjectivity of initial parameter selection, the dependence of feature weights setting on experience, and the bad adaptability of multidimension feature node analysis. At the data processing layer of the algorithm, the parameter of the model and observation sequence are optimized by the method of clustering; and then at the training layer of the algorithm, the single-feature Hidden Markov Model is used to model the multi-feature Hidden Markov Model; finally, an adaptive genetic algorithm based on the information gain is used to optimize and reduce the state sequence generated by the Hidden Markov Model. The problems of feature weight setting and parameter initial value selection are solved effectively. Experimental results show that the proposed algorithm effectively improves the accuracy of the high-dimensional health state of large-scale edge layer nodes compared with the existing algorithms.
The existing Siamese object tracking algorithms easily lead to tracking drift under the influence of object deformation and occlusion, this paper proposes an improved object tracking algorithm based on deep contour extraction networks to achieve stable detection and tracking of any object under complex backgrounds. First, the contour detection network automatically obtains the closed contour information on the object and uses the flood-filling clustering algorithm to obtain the contour template. Then, the contour template and the search area are input into the improved Siamese network so as to obtain the optimal tracking score value and adaptively update the contour template. If the object is fully obscured or lost, the Yolov3 network is used to search the object in the entire field of view to achieve stable tracking throughout the process. A large number of qualitative and quantitative simulation results show that the improved model can not only improve the object tracking performance under complex backgrounds, but also improve the response time of airborne systems, which is suitable for engineering applications.
With the application of artificial intelligence on the embedded platform, the k-means clustering algorithm, as the basis of the artificial intelligence method, is implemented on the embedded platform. Energy consumption is the key for the algorithm implementation on the embedded platform. In order to reduce the energy consumption of the k-means on the embedded platform, an approximate computing method based on cross-layer dynamic precision scaling for the k-means is proposed. First, the iteration process is constrained from the distance between data point to centroid and data point change trend. And a dynamic precision scaling method is proposed. Then the data reorganization and access method of external memory is designed from the structural level, which can realize the access of approximate memory. In addition, the approximate adder and multiplier are designed which can automatically adjust the calculation accuracy. Finally, the approximate computing of the k-means is realized. Experimental results show that the proposed method can reduce the energy consumption by 55%~58% compared with the accurate computing without affecting the quality of clustering. The proportion of the energy saving is the highest.
In order to improve the network recognition accuracy in the low complexity condition, a piecewise linear sigmoid function approximation based on the distribution probability of the neurons’ values is proposed only with one addition circuit. The sigmoid function is first divided into three fixed regions. Second, according to the neurons’ values distribution probability, the curve in each region is segmented into sub-regions to reduce the approximation error and improve the recognition accuracy. The slope of the piecewise linear function is set as 2-n, effectively reducing the hardware implementation complexity. Experiments performed on Xilinx’s FPGA-XC7A200T implement the MNIST handwritten digits recognition. The results show that the proposed method achieves a 97.45% recognition accuracy in a deep neural network and 98.42% in a convolutional neural network, up to 0.84% and 0.57% higher than other approximation methods only with one addition circuit.
In order to improve the recognition ability of noisy images, a method of contour reconstruction based on depth residuals learning is proposed. The sharpening template matching technique is used to enhance the noisy image information, the local gray level information on the image is used to construct the edge active contour model of the image, and the active contour lasso method is used to reconstruct the image with a high resolution. The feature quantities of local gray energy and local gradient energy of the noisy image are extracted, and a convolutional neural network classifier is constructed to classify the features. The learning depth of the learning convolutional neural network is judged by combining the similarity of the gray histogram of the image. The resolution ability of image detail information is improved, and the contour high resolution reconstruction of the noisy image is realized. Simulation results show that the proposed method has a high resolution and a high peak signal to noise ratio (PSNR), which improves the recognition ability of the image effectively.
Aiming at the uncertainty of the orbit status of non-cooperative space objects, a dynamic inference method for the orbit status of space objects based on dynamic Bayesian networks is proposed. First, the semantic model for the orbit status of space objects is established, and the semantic relationships such as the orbit status, orbit class and orbit change are explained. Second, the orbit status characteristics are analyzed, and the hierarchical division method for coplanar and noncoplanar orbit change is constructed. Then, based on the dynamic Bayesian network, an inference method for the orbit status of space objects is established, and the relationships between orbit status, orbit class, and orbit change are used to obtain the dynamic change process of the orbit status. Finally, the proposed method is validated by comparing with actual situations of space objects of different orbital classes. Experimental results demonstrate that the proposed dynamic inference method for the orbit status of space objects can inference the orbit status with uncertainty and obtain the change process, which provides support and assistance for further decision-making.
In order to effectively allocate the idle spectrum and improve spectrum utilization of cognitive wireless sensor networks, it is necessary to design an efficient spectrum allocation algorithm. Aiming at the problem of spectrum allocation in cognitive wireless sensor networks, an improved method for spectrum allocation is suggested. A new chaotic dynamic clonal evolution algorithm is designed. Then the graph theory coloring model is established with the corresponding fitness function derived. Traditional evolutionary algorithms have the problem of premature convergence, so chaotic operators, adaptive operators and cloning operators are added to the traditional evolutionary algorithms to accelerate the convergence of the algorithm. The chaotic dynamic clonal evolutionary algorithm is compared with the simulated annealing algorithm and the ant colony algorithm by simulation. The simulation results show that compared with the ant colony algorithm and the simulated annealing algorithm, the chaotic dynamic clonal evolution algorithm can effectively improve the global search ability, and significantly improve the network benefit value of spectrum allocation. The results also show that the proposed chaotic dynamic clonal evolution algorithm can make full use of existing spectrum resources and improve the system throughput.
With the rapid development of wireless communication technology, how to use an antenna to realize the function of multiple antennas has become a hot research topic. The double circularly polarized antenna studied in this paper can achieve right-handed circular polarization and left-handed circular polarization in the same frequency band by using a circular annular groove and two orthogonal L-shaped feed lines in the ground plane. At the same time, properly grooving at the center of the ground plane and adding parasitic elements to the front of the L-shaped feed line can change the current on the surface of the feed line and in the ground plane, thereby improving the circular polarization performance of the antenna. Measurement and simulation results show that the impedance bandwidth of the antenna is about 59% (3.13~5.75GHz), and the 3dB axial ratio bandwidth is about 40.5% (3.23~4.87GHz). Also, the isolation between the two ports is higher than 10dB. The results show that the designed antenna has a good performance in both impedance bandwidth and axial ratio bandwidth.
In order to design an optical fiber sensor with a simple structure which can simultaneously measure temperature and relative humidity, we use the multimode fiber cascade Bragg grating to form the basic structure of the sensor. First, a 6mm multimode fiber is connected to a Bragg grating. Then the diameter of the multimode fiber is etched to 40μm by hydrofluoric acid. Finally, a layer of carboxymethyl cellulose hydrogel is coated on the multimode fiber. The temperature and humidity response of the fabricated optical fiber sensor is tested. Experimental results show that the designed sensor has a humidity sensitivity of 69.6pm/% RH and a temperature sensitivity of 15pm/℃. The designed sensor is very sensitive to temperature and humidity, and has a good application prospect.
To solve the problem that the traditional automatic text summary model cannot generate a high-quality long text summary due to the limitation of the length of the RNN (Recurrent Neural Network), a model of abstractive text summarization for topic-aware communicating agents has been proposed. First, the problem that the LSTM (Long Short-Term Memory) input sequence is too long to generate the abstract with prior information has been solved by dividing the encoder into multiple collaborating agents. Then for providing topic information and improving the correlation between the generated abstract and the source text, the joint attention mechanism has been added into our model. Finally, a hybrid training method with reinforcement learning has been employed in order to solve the problem of exposure bias, and optimize the evaluation index directly. The results show that our model not only generate long text summaries with prominent themes, but also has a higher score than the state-of-the-art models, which indicates that with the help of topic information, the model for communicating agents can be expected to generate long text summaries better.
To overcome the shortcomings of traditional micro-motion classification of spatial cone targets, such as manual construction, feature extraction, and lack of generality, intelligence and poor classification performance under strong noise, a new network model combining a convolutional neural network and a bidirectional long short-term memory network is proposed. The network combines the residual network, inception network and bidirectional long short-term memory network into an integrated network. By increasing the depth and width of the network to mine the abstract features of higher dimensions, the classification accuracy of the network can be improved. The reasoning ability of the bidirectional long short-term memory network can improve the fault tolerance of the network, and the advantages of time series classification and the jumping bypass branch structure of the residual network can also reduce parameter redundancy and speed up network training. Simulation results show that the network model not only achieves faster intelligent classification, but also improves the accuracy of ResNet-18 and GoogLeNet models by 5% and 4% respectively, thus verifying the validity and generalization ability of the model.
An algorithm for routing optimization of an energy and path constrained wireless sensor network is proposed to solve the energy limitation problem which is caused by the frequent use of a single path by traditional wireless sensor network routing algorithms. By considering the load balancing and energy efficiency of wireless sensor networks, the concept of flight feasible domain is introduced to achieve efficient data transmission. Furthermore, energy and distance factors are added to ensure uniform and reasonable energy distribution among network nodes, so that the low-power and energy-efficient functional requirements of the wireless sensor network are satisfied. . Experimental results show that the proposed method can improve the network load balancing effectively, avoid the network segmentation caused by premature exhaustion of some nodes, and prolong the lifetime of the network.
In order to improve the security transmission performance of dual-polarized satellites, a constellation splitting security transmission system based on the multi-parameter weighted-type Fractional Fourier Transform (MP-WFRFT) is proposed. By analyzing the time-frequency characteristics of signals, the constellation splitting function is constructed and solved by the Genetic Algorithm (GA). The constellation splitting criteria of the MP-WFRFT in the two-dimensional surface and three-dimensional sphere are mainly explored. Simulation results show that the splitting pattern is determined by the ratio of time-domain terms on the MP-WFRFT and the transmitting signal can be camouflaged by setting different time-domain terms ratio constrains to prevent the eavesdropper from acquiring the correct message. The error in the transform parameter can improve the security performance. The larger the parameter error of the MP-WFRFT is, the higher secrecy capacity can be achieved, which verifies the rationality and effectiveness of the proposed scheme.
Spatially coupled low density parity check (SC-LDPC) codes can achieve a better decoding performance with a small message recovery latency due to the sliding window decoding. An improved decoding scheme based on window extension is proposed for further enhancing the performance of the sliding window decoding. In contrast to conventional sliding window decoding, the window size of this scheme can vary according to the average logarithmic likelihood ratio (LLR) value of the target symbol. Specifically, for every iteration in the decoding process, we compare the average LLR value of the target symbol with the preset threshold. If the average LLR value of the target symbol is less than the preset threshold and the current window size does not exceed the maximum size, the decoding window size adds one and then performs a new iteration with the new window size. By this means, the scheme can achieve trade-off between decoding performance, complexity and latency. Simulation results on the additive white Gaussian noise (AWGN) channel show that this scheme can significantly improve the sliding window decoding performance of SC-LDPC codes.