Thomas Lu

Optical N4 implementation of a two-dimensional wavelet transform

A two-dimensional wavelet transform is implemented by a bank of wavelet transform filters in the Fourier domain. An optical N 4 multichannel correlator architecture is proposed to perform parallel optical 2-D wavelet transforms. A holographic recording scheme is proposed to implement such a wavelet transform filter array. The optical experimental results are presented using the computer-generated transmittance masks as the wavelet transform filters.

Evaluation of holographic subsurface radar for NDE of space shuttle thermal protection tiles

Experiments have been carried out to evaluate holographic subsurface radar (RASCAN) for non-destructive evaluation (NDE) of subnominal bond conditions between the Space Shuttle Thermal Protection System tiles and the aluminum substrate. Initial results have shown detection of small voids and spots of moisture between Space Shuttle thermal protection tiles and underlying aluminum substrate. The characteristic feature of this device is the ability to obtain one-sided radar soundings/images with high sensitivity (detecting of wire of 20 micron and less in diameter), and high resolution (2 cm lateral resolution) in the frequency band of 3.6-4.0 GHz. JPLs advanced high-speed image processing and pattern recognition algorithms can be used to process the data generated by the holographic radar and automatically detect and measure the defects. Combining JPLs technologies with the briefcase size, portable RASCAN system will produce a simple and fully automated scanner capable of inspecting dielectric heat shielding materials or other spacecraft structures for cracks, voids, inclusions, delamination, debonding, etc.. We believe this technology holds promise to significantly enhance the safety of the Space Shuttle and the future CEV and other space exploration missions.

Optimization of OT-MACH Filter Generation for Target Recognition

An automatic Optimum Trade-off Maximum Average Correlation Height (OT-MACH) filter generator for use in a gray-scale optical correlator (GOC) has been developed for improved target detection at JPL. While the OT-MACH filter has been shown to be an optimal filter for target detection, actually solving for the optimum is too computationally intensive for multiple targets. Instead, an adaptive step gradient descent method was tested to iteratively optimize the three OT-MACH parameters, α, β, and γ. The feedback for the gradient descent method was a composite of the performance measures, correlation peak height and peak to side lobe ratio. The automated method generated and tested multiple filters in order to approach the optimal filter quicker and more reliably than the current manual method. Initial usage and testing has shown preliminary success at finding an approximation of the optimal filter, in terms of α, β, γ values. This corresponded to a substantial improvement in detection performance where the true positive rate increased for the same average false positives per image.

Integration of the reconfigurable self-healing eDNA architecture in an embedded system

In this work we describe the first real world case study for the self-healing eDNA (electronic DNA) architecture by implementing the control and data processing of a Fourier Transform Spectrometer (FTS) on an eDNA prototype. For this purpose the eDNA prototype has been ported from a Xilinx Virtex 5 FPGA to an embedded system consisting of a PowerPC and a Xilinx Virtex 5 FPGA. The FTS instrument features a novel liquid crystal waveguide, which consequently eliminates all moving parts from the instrument. The addition of the eDNA architecture to do the control and data processing has resulted in a highly fault-tolerant FTS instrument. The case study has shown that the early stage prototype of the autonomous self-healing eDNA architecture is expensive in terms of execution time.

Neural network post-processing of grayscale optical correlator

In real-world pattern recognition applications, multiple correlation filters can be synthesized to recognize broad variation of object classes, viewing angles, scale changes, and background clutters. Composite filters are used to reduce the number of filters needed for a particular target recognition task. Conventionally, the correlation peak is thresholded to determine if a target is present. Due to the complexity of the objects and the unpredictability of the environment, false positive or false negative identification often occur. In this paper we present the use of a radial basis function neural network (RBFNN) as a post-processor to assist the optical correlator to identify the objects and to reject false alarms. Image plane features near the correlation peaks are extracted and fed to the neural network for analysis. The approach is capable of handling large number of object variations and filter sets. Preliminary experimental results are presented and the performance is analyzed.

Development of streamlined OT-MACH-based ATR algorithm for grayscale optical correlator

JPL is developing an Advanced Autonomous Target Recognition (AATR) technology to significantly reduce broad area search workload for imagery analysts. One of the algorithms to be delivered, as part of JPL ATR Development and Evaluation (JADE) project, is the OT-MACH based ATR algorithm software package for grayscale optical correlator. In this paper we describe the basic features and functions of the software package as currently implemented. Automation of filter synthesis and test for GOC, particularly the automation of OT-MACH parameter optimization, is discussed.

Neural Network Target Identification System for False Alarm Reduction

A multi-stage automated target recognition (ATR) system has been designed to perform computer vision tasks with adequate proficiency in mimicking human vision. The system is able to detect, identify, and track targets of interest. Potential regions of interest (ROIs) are first identified by the detection stage using an Optimum Trade-off Maximum Average Correlation Height (OT-MACH) filter combined with a wavelet transform. False positives are then eliminated by the verification stage using feature extraction methods in conjunction with neural networks. Feature extraction transforms the ROIs using filtering and binning algorithms to create feature vectors. A feed forward back propagation neural network (NN) is then trained to classify each feature vector and remove false positives. This paper discusses the test of the system performance and parameter optimizations process which adapts the system to various targets and datasets. The test results show that the system was successful in substantially reducing the false positive rate when tested on a sonar image dataset.

Holographic Subsurface Radar as a Device for NDT of Construction Materials and Structures

A relatively rare type of subsurface radar – holographic radar – is described in this paper as a tool for non-destructive testing (NDT) of construction materials and structures. Its principle of operation, advantages and disadvantages are considered. Holographic subsurface radar, operating several discrete frequencies, is used to illuminate a sufficiently extensive area of a surface of opaque dialectical medium to be inspected to register interference between reflected from objects and reference waves. In a lossy media with low level of microwaves attenuation, reconstruction algorithms could be applied for obtaining the subsurface image in such a manner as in optical holography. An attempt is made to highlight significant application areas and problem cases where this type of radar could potentially be applied as a device for NDT of construction materials and structures. The paper describes results of different building surveying including objects of historical heritage. Space shuttle thermal protection system tiles were investigated in some other experiments. Each application area is illustrated by relevant data acquired in laboratory experiments or field tests.

Composite wavelet filters for enhanced automated target recognition

Automated Target Recognition (ATR) systems aim to automate target detection, recognition, and tracking. The current project applies a JPL ATR system to low-resolution sonar and camera videos taken from unmanned vehicles. These sonar images are inherently noisy and difficult to interpret, and pictures taken underwater were unreliable due to murkiness and inconsistent lighting. The ATR system breaks target recognition into three stages: 1) Videos of both sonar and camera footage are broken into frames and preprocessed to enhance images and detect Regions of Interest (ROIs). 2) Features are extracted from these ROIs in preparation for classification. 3) ROIs are classified as true or false positives using a standard Neural Network based on the extracted features. Several preprocessing, feature extraction, and training methods are tested and discussed in this report.

Feature Extraction and Selection Strategies for Automated Target Recognition

Several feature extraction and selection methods for an existing automatic target recognition (ATR) system using JPLs Grayscale Optical Correlator (GOC) and Optimal Trade-Off Maximum Average Correlation Height (OT-MACH) filter were tested using MATLAB. The ATR system is composed of three stages: a cursory regionof- interest (ROI) search using the GOC and OT-MACH filter, a feature extraction and selection stage, and a final classification stage. Feature extraction and selection concerns transforming potential target data into more useful forms as well as selecting important subsets of that data which may aide in detection and classification. The strategies tested were built around two popular extraction methods: Principal Component Analysis (PCA) and Independent Component Analysis (ICA). Performance was measured based on the classification accuracy and free-response receiver operating characteristic (FROC) output of a support vector machine(SVM) and a neural net (NN) classifier.