David de la Mata-Moya

Spatial-Range Mean-Shift Filtering and Segmentation Applied to SAR Images

The mean-shift (MS) algorithm is applied for reducing speckle noise and segmenting synthetic aperture radar (SAR) images. Two coastal images acquired by Envisats advanced SAR (ASAR) [European Space Agency (ESA)] are used. Studies of the MS parameters are carried out according to the desired product: a speckle filtered image where textures and edges are preserved, or a segmented image, where land and sea are distinguished, as a previous stage for obtaining a land mask and detecting the coastal line. In all cases, Gaussian kernels are used. Speckle filtering results are compared with those obtained using uniform kernels, proving that the former provides better results than the latter. A segmentation approach based on the positions and frequencies at which the MS converges is applied. The use of a combined spatial-range processing and the corresponding bandwidths makes the MS suitable for the two proposed problems. The solid theoretical basis of this procedure allows designing a guided search of the best parameters according to the desired solution, avoiding a tedious trial-and-error process. Although the used images have different characteristics, results prove that similar sets of parameters can be used, showing some degree of robustness with respect to the image, for a given sensor and image acquisition mode.

Combining MLPs and RBFNNs to Detect Signals With Unknown Parameters

The detection of Gaussian signals with an unknown correlation coefficient rhos is considered. Solutions based on neural networks (NNs) are studied, and a strategy for designing committee machines in a composite hypothesis test is proposed. A single multilayer perceptron (MLP) has been trained with rhos uniformly varying in [0, 1]. Considering the decision boundaries for rhos = 0 and rhos = 1 and how an MLP approximates them, a detection scheme composed of two MLPs has been proposed. One of them MLP1 has been trained with rhos uniformly varying in [0, 0.5], and the other one MLP2 has been trained with rhos uniformly varying in [0.5, 1]. For making a decision, the higher output is compared to a threshold for each false-alarm probability (P FA). This strategy simplifies the task of finding a compromise solution between the computational cost and the approximation error and outperforms the single-MLP detector. When MLP1 is substituted with a radial basis function NN (RBFNN), a new combination strategy of the outputs is required. We propose separately thresholding the outputs and applying them to an or logic function. The performance of this detector is slightly better than the two-MLP one, and the computational cost is significantly reduced.

Approximating the Neyman-Pearson detector for swerling I targets with low complexity neural networks

This paper deals with the application of neural networks to approximate the Neyman-Pearson detector. The detection of Swerling I targets in white gaussian noise is considered. For this case, the optimum detector and the optimum decision boundaries are calculated. Results prove that the optimum detector is independent on TSNR, so, under good training conditions, neural network performance should be independent of it. We have demonstrated that the minimum number of hidden units required for enclosing the optimum decision boundaries is three. This result allows to evaluate the influence of the training algorithm. Results demonstrate that the LM algorithm is capable of finding excellent solutions for MLPs with only 4 hidden units, while the BP algorithm best results are obtained with 32 or more hidden units, and are worse than those obtained with the LM algorithm and 4 hidden units.

Statistical Analysis of SAR Sea Clutter for Classification Purposes

Statistical analysis of radar clutter has always been one of the topics, where more effort has been put in the last few decades. These studies were usually focused on finding the statistical models that better fitted the clutter distribution; however, the goal of this work is not the modeling of the clutter, but the study of the suitability of the statistical parameters to carry out a sea state classification. In order to achieve this objective and provide some relevance to this study, an important set of maritime and coastal Synthetic Aperture Radar data is considered. Due to the nature of the acquisition of data by SAR sensors, speckle noise is inherent to these data, and a specific study of how this noise affects the clutter distribution is also performed in this work. In pursuit of a sense of wholeness, a thorough study of the most suitable statistical parameters, as well as the most adequate classifier is carried out, achieving excellent results in terms of classification success rates. These concluding results confirm that a sea state classification is not only viable, but also successful using statistical parameters different from those of the best modeling distribution and applying a speckle filter, which allows a better characterization of the parameters used to distinguish between different sea states.

MLP-CFAR for improving coherent radar detectors robustness in variable scenarios

Neural Networks based CFAR techniques are proposed in an improved coherent radar detector.A coherent detector using a unique CFAR is compared to the classical bank of CFAR techniques.The filter bank output is statistically analyzed to prove Gaussian CFARs unfeasibility.The proposed neural CFAR can be applied to any clutter distribution or detection strategy.A comparative study is carried out on a simulated scenario with complex target trajectories. This paper tackles the detection of radar targets with unknown Doppler shift in presence of clutter. A Neural Network based Constant False Alarm Rate (CFAR) technique is proposed for adapting the detection threshold in an improved architecture based on the Generalized Likelihood Ratio (GLR) detector. Detection schemes based on Doppler processors (Moving Target Indicator (MTI) and Moving Target Detector (MTD)) and conventional CFAR detectors are considered as reference. In these reference solutions, interference is assumed Gaussian and white at the output of each Doppler filter, so conventional incoherent CFAR detectors are applied. The outputs of the CFAR detectors are combined using an OR operation to decide the presence of a target if, at least, one of the CFARs declares it. As a result, the P FA is higher than the desired one, as we prove. In this paper, an improved detector is presented that combines the following features: a better approximation to the Neyman-Pearson detector based on the GLR (selecting the maximum filter bank output), and a unique CFAR detector applied to the squared modulus of the maximum filter bank output. Due to the non-linear nature of the maximum function, conventional CFAR detectors are not suitable. A Neural Network CFAR solution is proposed. A general design method is presented. Results prove that the designed CFAR allows the exploitation of the better detection capabilities of the detector based on the maximum function, providing a higher probability of detection while fulfilling the probability of false alarm requirement. The proposed method can be extended to other detection strategies and radar scenarios.

Neural network based solutions for ship detection in SAR images

Ship detection is nowadays quite an important issue in tasks related to sea traffic control, fishery management and ship search and rescue. Although it has traditionally been carried out by patrol ships or aircrafts, coverage and weather conditions can become a problem. Synthetic aperture radars can surpass these coverage limitations and work under any climatological condition. Two ship detectors are proposed in this paper. The first one is a MLP-based detector that uses K-distribution parameters to characterize the sea clutter and the brightness of the pixels to detect ships. The second one is the double parameters model, DPM, proposed in the literature. While the DPM-based detector gives rise to some false alarms, leading to the need of a discrimination stage, and has some troubles when the ships are in rough water, the MLP-based detector along with the combination of both sea and ship features obtains better results in terms of detection and false alarm rates.

High order neural network based solution for approximating the Average Likelihood Ratio

The detection of gaussian signals with unknown correlation coefficient, ρs is considered. A strategy for designing high order neural networks (HONN) in composite hypothesis test is proposed. A HONN trained with ρs varying uniformly in [0, 1] is considered to approximate the Average Likelihood Ratio (ALR). In order to compare the suitability of the approximation, a sub-optimal solution based on constrained generalized likelihood ratio is used. A study of the computational cost is carried out. Results show that a HONN is able to approximate the ALR with a low computational cost.

Coherent Detection of Swerling 0 Targets in Sea-Ice Weibull-Distributed Clutter Using Neural Networks

The detection of Swerling 0 targets in movement in sea-ice Weibull-distributed clutter by neural networks (NNs) is presented in this paper. Synthetic data generated for typical sea-ice Weibull parameters reported in the literature are used. Due to the capability of NNs for learning the statistical properties of the clutter and target signals during a supervised training, high clutter reduction rates are achieved, reverting on high detection performances. The proposed NN-based detector is compared with a reference detector proposed in the literature that approximates the Neyman-Pearson (NP) detector. The results presented in the paper allow empirically demonstrating how the NN-based detector outperforms the detector taken as reference in all the cases under study. It is achieved not only in performance but also in robustness with respect to changes in sea-ice Weibull-distributed clutter conditions. Moreover, the computational cost of the NN-based detector is very low, involving high signal processing speed.

Detection of known targets in Weibull clutter based on Neural Networks. Robustness study against target parameters changes

The coherent detection of targets in presence of clutter and noise is considered in this study. Several clutter models are proposed in the literature, although the commonly used for sea and land clutter returns is the Weibull one. Our case of study involves that the target is known a priori, the clutter is Weibull-distributed and a white Gaussian noise is present. In this case, obtaining analytical expressions for the optimum detector is very difficult, so suboptimum solutions are taken as reference. One of this solutions is the target sequence known a priori (TSKAP) detector. This detector has several problems because it is designed for specific target and clutter parameters. So, in order to reduce these problems, a new solution is proposed, which is based in neural networks (NNs). The NNs selected are the MultiLayer Perceptrons (MLPs), which are able to learn from different environments. But, what does it happen if the radar (target or clutter) testing conditions are different of the design ones? In this case, a robustness study with respect to the target Doppler frequency is done for different radar conditions, which shows that the behavior of the proposed solution against this changes is better than the detector taken as reference, the TSKAP detector.

Radar detection with the Neyman–Pearson criterion using supervised-learning-machines trained with the cross-entropy error

The application of supervised learning machines trained to minimize the Cross-Entropy error to radar detection is explored in this article. The detector is implemented with a learning machine that implements a discriminant function, which output is compared to a threshold selected to fix a desired probability of false alarm. The study is based on the calculation of the function the learning machine approximates to during training, and the application of a sufficient condition for a discriminant function to be used to approximate the optimum Neyman–Pearson (NP) detector. In this article, the function a supervised learning machine approximates to after being trained to minimize the Cross-Entropy error is obtained. This discriminant function can be used to implement the NP detector, which maximizes the probability of detection, maintaining the probability of false alarm below or equal to a predefined value. Some experiments about signal detection using neural networks are also presented to test the validity of the study.