Figure 2 - uploaded by José luis Pasciaroni
Content may be subject to copyright.
Source publication
del Sur -Av. Alem 1253 -(8000) Bahía Blanca -Bs.As. Abstract— The Neural Network Cell Average -Order Statistics Constant False Alarm Rate (NNCAOS CFAR) detector is presented in this work. NNCAOS CFAR is a combined detection methodol-ogy which uses the effectiveness of neural networks to search for non homogeneities like clutter banks and multiple t...
Contexts in source publication
Context 1
... fa M M ! k OS ! G OS M M k ! ! (6) Figure 2, shows the OS CFAR threshold multiplier į OS variation against the k sample for M =40, P fa =1 u 10 -3 and P fa =1 u 10 -5 for a Weibull background with parameters Į =1 and ȕ =2. The curves were obtained by Monte Carlo simulation method. The results are very similar to those obtained by Eq. 6. The principal drawback of applying Eq. 6 is that only integer values of į OS should be used due to the factorial function, making it difficult to obtain results for intermediate values. Additional motivations for developing a new CFAR method and a detailed description of the novel Neural Alarm Rate (NNCAOS CFAR) detector are presented in the following. Excessive number of false alarms at clutter edges and degradation of the P d in multiple target environments in the CA CFAR detector are the prime motivations for exploring other CFAR schemes that discriminate between interference and the primary targets (Gandhi and Kassam, 1988). In the case of non homogeneous radar returns (with CB or MT), the CA CFAR P fa is no longer constant. The samples corresponding to the CB and MT, are aver- aged in the CFAR window, resulting an overall threshold increase and false detections at clutter edges. On the other hand, OS CFAR detector is more robust than CA CFAR, especially when the k sample is high (near M ). However, missing detection could occur in MT situations. The performance of this processor is highly dependent upon the values for k . Despite that OS CFAR detector exhibits some loss of detection performance in homogeneous noise background compared with CA CFAR, its performance in a multiple target environment is clearly superior (Gandhi and Kassam, 1988). We take into account that the performance of the OS CFAR processor is highly dependent upon the values for k . For example, if a single extraneous target appears in the reference window of appreciable magnitude, it occupies the highest ranked cell with high probability. The estimate will almost always set the threshold based on the value of the interfering target. This results in an increase of the overall threshold and may lead to a target miss. If, on the other hand, k is chosen to be less than the maximum value, the OS CFAR processor will be in- fluenced only slightly for up to M-k interfering targets (Gandhi and Kassam, 1988). Of course, we are interested in both cases, i.e., regu- lating the false alarm at the clutter edges and minimiz- ing target miss in the multiple target situations (Gandhi and Kassam, 1988). For this reason, different CFAR processors have been proposed to overcome the variety of situations that could be present within the radar return, i.e., CB, jamming, ice, multiple target, etc. (Gandhi and Kassam, 1988; Rohling, 1983; Haykin et al. , 1991; Haykin and Deng, 1991; Smith and Varshney, 2000). If it were possible to have a priori knowledge of each situation, the most appropriate CFAR processor could be chosen for each case in order to achieve the best P d while at the same time, to maintain intact the CFAR function that is to keep a constant false alarm rate. Several authors have proposed di ff erent analytical methods to solve non homogeneous situations within the radar return (Smith and Varshney, 2000; Doyuran and Tanik, 2007). In this work, neural networks are proposed to take advantage of their efficiency (in terms of lower computation time than other schemes) considering real time processing (Gálvez et al. , 2011). Using NN it is possible to make a classification to each radar return while working in real time. Haykin et al. (1991) and Haykin and Deng (1991) proposed a clutter classification methodology to distinguish between several mayor classes of radar returns including weather, birds and aircraft. Of particular interest is the use of a multi- layer feedforward NN as the basis for classifying primary radar returns in, for example, aid traffic control environment applications (Haykin et al. , 1991; Haykin and Deng, 1991). CANN CFAR, presented in a former work, com- bines Maximum Likelihood (ML) for clutter parameter estimation, and NN for radar return homogeneity testing and clutter bank (transition points) and size estimation. It was demonstrated that this detector has better performance than conventional CFAR processors, especially when detecting targets located near clutter non homogeneities (Gálvez et al. , 2011). An additional advantage of that technique is its efficiency when determining clutter transition points, bank size and threshold setting (Gálvez et al. , 2011). On the other hand, it exhibits difficulties when discriminating multiple target, since it confuses multiple target situations with non homogeneities, diminishing in these cases the probability of detection considerability. The NNCAOS CFAR is developed in this work with the purpose of improving the discussed difficulties. NN are used with the purpose of making a priori analysis to the radar return before an appropriate CFAR detector ...
Context 2
... NNCAOS CFAR is a composed process, which takes advantage of NN as applied to CA CFAR and/or OS CFAR concepts. Figure 3 depicts the NNCAOS CFAR block diagram. Like CANN CFAR, radar return enters to the Maximum Likelihood (ML) parameter estimation block where clutter parameters are estimated. We choose a relatively large number of samples (2 M ) at the end of the radar return for this estimation, assuming homogeneity. Then, we calculate the threshold multiplier for the CA CFAR process, į CA ( P fa ,M ), according to Gálvez et al. (2011), and use its value in the thresholding blocks. In the case of the OS CFAR processors, we obtain the threshold multiplier į OS ( P fa ,M ) from Fig. 2 for the corresponding k th sample and P fa , assuming homogeneous clutter. Two NN blocks, NN 1 and NN 2 , analyze groups of M samples ( M =40). NN 1 block searches for CB and NN 2 looks for MT. As a result, each sample is labeled as homogeneous, CB or MT. The algorithm selects a CA CFAR in the case of homogeneous radar return, an OS CFAR processor with a high k th sample (the OS 35 . i.e., k =35 for the case of a CB, or the OS 20 for the case of a MT). Then, the most appropriate CFAR processor is applied to each sample in the CUT, and every CFAR processor obtains the threshold T , according to Eq. 4 and Eq. 5. Finally, the detection is carried out, a target is de- clared if the signal amplitude in the CUT is greater than the threshold. We obtain thus, a robust system, which maintains a constant probability of false alarm P fa even for non homogeneous returns while at the same time, it achieves a higher (or equal) probability of detection P d than the classical systems in most of the cases. In this section, NNCAOS CFAR simulation results are shown in order to illustrate and discuss its performance. NN training is performed by means of the backpropagation algorithm considering Weibull distributed radar returns. In the case of NN 1 homogeneity test block, a network composed by 40 inputs, 40 neurons in its hidden layer and only one output was trained for 20000 epochs by means of 6160 radar returns; 400 samples of homogeneous radar returns with diverse parameters ( Į =1; ȕ =2, 1.6, 1.4, 1.33), and 5760 samples with non homogeneous returns containing different size and parameter clutter banks situated at several positions (Gálvez et al. , 2011). The NN 2 multiple target test block, is also a neural network. This block with 40 inputs, 40 neurons in its hidden layer and only one output, was trained for 15060 epochs by means of 6655 homogeneous radar returns samples (without any target) and 6655 radar returns samples containing multiple targets situated at different positions within the CFAR window. NNCAOS CFAR performance is compared to the CA, OS 20 , the OS 35 and the CANN CFAR detectors. The results are illustrated in different figures. In these figures, we represent the NNCAOS CFAR with a right line, the CA CFAR with asterisks (*), the OS CFAR with a k th sample of 20 (OS 20 ) with a plus sign (+), the OS CFAR with a k th sample of 35 (OS 35 ) with a triangle ( ̈ ) and the CANN CFAR with a circle (o). NNCAOS CFAR performance is compared to the other detectors for different representative radar return ...
Citations
... Cheikh and Soltani [25] studied the problem of radar target detection with jamming background obeying the K distribution by using a multi-layer perceptual network (MLP) and a radial basis function. Gálvez et al. [26] proposed the neural network cell average order statistics CFAR (NNCAOS-CFAR) by combining a maximum likelihood-based CFAR (ML-CFAR) with a neural network. The detector can adaptively select the CA-CFAR and OS-CFAR to detect targets according to the clutter background, but it is limited to a uniform clutter background. ...
Abstract With oceanic reverberation and a large amount of data being the main sources of interference for underwater acoustic target detection, it is difficult to obtain a more robust detection performance by relying on the traditional constant false alarm rate (CFAR) detection method. An adaptive sonar CFAR detection method based on a back propagation (BP) neural network is proposed. The method combines the artificial intelligence algorithm and the traditional detection algorithm, and uses the classification ability of the algorithm to select the detection algorithm, which can effectively improve the adaptation ability of the algorithm and the environment and the false alarm control ability. The method combines the artificial intelligence algorithm and the traditional detection algorithm, and uses the classification ability of the algorithm to select the detection algorithm, which can effectively improve the adaptation ability of the algorithm and the environment and the false alarm control ability. This method uses a BP neural network to train the target echo signal to complete the clutter background classification and establish the clutter background recognition classification set. According to the output result of each classification, the best CFAR detector is selected from four CA/SO/GO/OS‐CFAR detectors to detect the target. The simulation results show the detection performance of the proposed method in a uniform environment, a multi‐target environment, and a clutter edge environment. The results show that the environment adaptability is strong for different clutter backgrounds, which further improves the control ability of false alarms under a non‐uniform background.
... Las investigaciones sobre el uso de las redes neuronales artificiales en la detección de blancos de radar comenzaron a aparecer con cierta frecuencia en la década de los 90 [198,[202][203][204][205]. Sin embargo es a partir del 2008 que la técnica comienza a ser realmente popular [49,50,122,135,174,197,[206][207][208][209][210][211][212][213], luego de que la tendencia se mantuviese estable entre los años 2000 y 2007 [134,[214][215][216]. ...
... Los resultados alcanzados ayudan a reforzar la aplicación de las redes neuronales a la solución de problemas de radar que ha sido abordada previamente en [49,50,197,212,[226][227][228]. Se considera que el método neuronal presentado puede extenderse a la selección de otras distribuciones relacionadas al clutter como son la Pareto [34,60], la Compuesta Gaussiana [236] y la K-K [35], entre otras. ...
... Para el diseño de la red neuronal, el autor tomó como punto de partida las soluciones dadas en [49,50,197,212,226] ...
La mayoría de los procesadores CFAR modifican el algoritmo de cálculo del promedio buscando un mejor desempeño frente a heterogeneidades. En cambio, no se establece ningún mecanismo para adaptar las soluciones a la fluctuación de la distribución estadística del clutter y su parámetro de forma.
Como respuesta a esta problemática, la presente investigación propone una variante de procesamiento CFAR que corrige continuamente el factor de ajuste, logrando así que la probabilidad de falsa alarma se mantenga estable alrededor del valor de diseño, aun cuando ocurran cambios en la estadística del clutter. La solución es novedosa tanto en su concepción como en las técnicas aplicadas. En primer lugar, la idea de corregir el factor de ajuste a partir de estimaciones de la estadística del clutter, y de aplicar fórmulas de transformación obtenidas por ajuste de curva, introduce una nueva metodología sin precedentes en la literatura. En segundo lugar, la propuesta devela una nueva aplicación de las redes neuronales, al emplearlas para la selección de la distribución preferencial del clutter y la estimación de su parámetro de forma. El diseño fue simulado en MATLAB empleando muestras generadas en computadora, y fue validado con lecturas tomadas de bases de datos internacionales.
... This research investigated the use of the neural network for detecting the homogeneity signals lead to the selection between CA-CFAR or OS-CFAR. The proposed scheme was capable of maintaining a constant probability of false alarm in various environments (Galvez et al., 2012). In 2015, the switching CA/OS CFAR using neural network was proposed. ...
... According to that detection result, the algorithm will select the best CFAR algorithm. The options of CFAR algorithms provided were CA-CFAR 32 cells, OS-CFAR 22 cells and OS-CFAR 31 cells [7]. Then in 2015, the switching algorithm between CA-CFAR and OS-CFAR has been designed and proposed using multi-layer perceptron network (MLPN). ...
Target detection is a mandatory task of radar system so that the radar system performance is mainly determined by its detection rate. Constant False Alarm Rate (CFAR) is a detection algorithm commonly used in radar systems. This method is divided into several approaches which have different performance in the different environments. Therefore, this paper proposes an ensemble neural network based classifier with a variation of hidden neuron number for classifying the radar environments. The result of this research will support the improvement of the performance of the target detection on the radar systems by developing such an adaptive CFAR. Multi-layer perceptron network (MLPN) with a single hidden layer is employed as the structure of base classifiers. The first step of this research is the evaluation of the hidden neuron number giving the highest accuracy of classification and the simplicity of computation. According to the result of this step, the three best structures are selected to build an ensemble classifier. On the ensemble structure, all of those three MLPN outputs then be collected and voted for getting the majority result in order to decide the final classification. The three possible radar environments investigated are homogeneous, multiple-targets and clutter boundary. According to the simulation results, the ensemble MLPN provides a higher detection rate than the conventional single MLPNs. Moreover, in the multiple-target and clutter boundary environments, the proposed method is able to show its highest performance.
... For the design of the neural network, the authors took as a start point the solutions given in [26,30,[39][40][41] for different situations. Consequently, the initial configuration of the network internal variables was the one presented in Table 1. ...
The main problem faced by naval radars is the elimination of the clutter input which is a distortion signal
appearing mixed with target reflections. Recently, the Pareto distribution has been related to sea clutter measurements suggesting that it may provide a better fit than other traditional distributions. The authors propose a new method for estimating the Pareto shape parameter based on artificial neural networks.
The solution achieves a precise estimation of the parameter, having a low computational cost, and outperforming the classic method which uses Maximum Likelihood Estimates (MLE). The presented
scheme contributes to the development of the NATE detector for Pareto clutter, which uses the knowledge of clutter statistics for improving the stability of the detection, among other applications.
... For the design of the neural scheme, the authors took as a start point the solutions given in [15,[33][34][35][36][37] for different situations. Consequently, the initial configuration of the neural network internal variables was arranged according to Table I. ...
The discrimination of the clutter interfering signal is a current problem in modern radars’ design, especially in coastal or offshore environments where the histogram of the background signal often displays heavy tails. The statistical characterization of this signal is very important for the cancellation of sea clutter, whose behavior obeys a K distribution according to the commonly accepted criterion. By using neural networks, the authors propose a new method for estimating the K shape parameter, demonstrating its superiority over the classic alternative based on the Method of Moments. Whereas both solutions have a similar performance when the entire range of possible values of the shape parameter is evaluated, the neuronal alternative achieves a much more accurate estimation for the lower Fig.s of the parameter. This is exactly the desired behavior because the best estimate occurs for the most aggressive states of sea clutter. The final design, reached by processing three different sets of computer generated K samples, used a total of nine neural networks whose contribution is synthesized in the final estimate, thus the solution can be interpreted as a deep learning approximation. The results are to be applied in the improvement of radar detectors, particularly for maintaining the operational false alarm probability close to the one conceived in the design.
... For the design of the neural scheme, the authors took as a start point the solutions given in [15,[33][34][35][36][37] for different situations. Consequently, the initial configuration of the neural network internal variables was arranged according to Table I. ...
The discrimination of the clutter interfering signal is a current problem in modern radarstextquoteright design, especially in coastal or offshore environments where the histogram of the background signal often displays heavy tails. The statistical characterization of this signal is very important for the cancellation of sea clutter, whose behavior obeys a K distribution according to the commonly accepted criterion. By using neural networks, the authors propose a new method for estimating the K shape parameter, demonstrating its superiority over the classic alternative based on the Method of Moments. Whereas both solutions have a similar performance when the entire range of possible values of the shape parameter is evaluated, the neuronal alternative achieves a much more accurate estimation for the lower Fig.s of the parameter. This is exactly the desired behavior because the best estimate occurs for the most aggressive states of sea clutter. The final design, reached by processing three different sets of computer generated K samples, used a total of nine neural networks whose contribution is synthesized in the final estimate, thus the solution can be interpreted as a deep learning approximation. The results are to be applied in the improvement of radar detectors, particularly for maintaining the operational false alarm probability close to the one conceived in the design.
... By this algorithm, the rate of SCR improvement was up to 12 dB. Then, the combination of ML-CFAR and neural network called NNCAOS-CFAR developed in 2012 [8]. This research investigated the use of neural network for detecting the homogeneity signals lead to the selection between CA-CFAR and OS-CFAR. ...
This paper presents the switching CA/OS CFAR using neural network for improving the radar target detection in non-homogeneous environment. This method uses one of between CA-CFAR and OS-CFAR as output threshold depends on the nearest value with the output of neural network. The neural network used in this research is the Multi-Layer Perceptron (MLP) consisted of two hidden layers. The input of neural network was as many as 3 consisted of CA and OS CFAR and Cell Under Test (CUT) value. The pattern of those inputs will be classified and recognized by the neural network by the training to calculate the preliminary threshold. That threshold will be compared to CA and OS CFAR to select the best final threshold. The method was examined with three simulated common radar cases including homogeneous background, multi target and clutter wall environment. The experiments show that the proposed method is capable to select properly based on the best performance of both CA and OS CFAR in homogeneous and non-homogeneous environments.
As one of the state-of-the-art technologies currently being studied in the field of target positioning, the entangled light quantum-based passive target positioning method uses quantum entangled states to detect and localize the target, thus providing highly-accurate and anti-interference positioning services. The conventional methods seldom consider the problem that maybe a pseudo-target is detected when there is actually no target due to the interference of particles, scatterers, and noise photons during light quantum transmission. To address this problem, this paper proposes a new passive target positioning method on uncertain detection path based on entangled light quantum. Specifically, first of all, by converting the continuous light energy into the discrete photon number, the photon return rate is used to calculate the number of arrival photons based on the counting characteristics of the single photon detector. Second, the detection probabilities of signal, idler, and noise photons are calculated to obtain the coincidence probability at the receiver. Third, according to the coincidence count value of the entangled light time pulse sequence, the Constant False Alarm Rate (CFAR) detection model is established, and the corresponding detection threshold is adaptively adjusted with the change of the background noise to achieve highly-sensitive target detection. Finally, the relative error for each light source is considered to select the optimal light sources with the smallest relative distance errors for target positioning. In addition, the results show that the detection probability of the proposed method is over 95% with the false alarm probability 0.05, the positioning error is centimeter-level, and the confidence probability within 2.60 cm can reach 92%.
Ship detection for SAR imagery plays a crucial rule in the field of remote-sensing image processing. Superpixel methods have attracted enormous interest in the recent years, resulting in superpixel-based CFAR (Constant False Alarm Rate) ship detection methods for SAR images become a hot research issue. However, existing methods take superpixel generation as a preprocessing, and CFAR detection is essentially carried out at the pixel level, and the influence of speckle noise has not been fundamentally overcome. This paper innovatively proposed a real superpixel-level CFAR detection method, which takes superpixels instead of pixels as the minimum processing primitives on the basis of reliable superpixel segmentation. Firstly, in order to obtain the background superpixel region of each under test superpixel more accurately and efficiently, we designed a ring topology, which was specially developed for non-Euclidean structure data like superpixels. Secondly, the Johnsonsu function is used to fit the superpixel clutters from the background ring, and the detection threshold was obtained by a given false alarm rate. Finally, the candidate targets were further optimized to carry out accurate detection results. Experimental results based on multiple sets of real SAR images shown that our method has significant advantages in terms of detection rate and false alarm rate compared to the traditional K-CFAR and the existing superpixel-based CFAR methods.
