Search

This paper presents an approach that combines conventional image processing with deep learning by fusing the features from the individual techniques. We hypothesize that the two techniques, with different error profiles, are synergistic. The conventional image processing arm uses three handcrafted biologically inspired image processing modules and one clinical information module. The image processing modules detect lesion features comparable to clinical dermoscopy information—atypical pigment network, color distribution, and blood vessels. The clinical module includes information submitted to the pathologist—patient age, gender, lesion location, size, and patient history. The deep learning arm utilizes knowledge transfer via a ResNet-50 network that is repurposed to predict the probability of melanoma classification. The classification scores of each individual module from both processing arms are then ensembled utilizing logistic regression to predict an overall melanoma probability. Using cross-validated results of melanoma classification measured by area under the receiver operator characteristic curve (AUC), classification accuracy of 0.94 was obtained for the fusion technique. In comparison, the ResNet-50 deep learning based classifier alone yields an AUC of 0.87 and conventional image processing based classifier yields an AUC of 0.90. Further study of fusion of conventional image processing techniques and deep learning is warranted. [ABSTRACT FROM AUTHOR]

A vision-based measurement approach for pill shape detection is presented along with other applications. Rapid and accurate pill identification is needed by medical and law enforcement personnel during emergencies. But real-world pill identification is challenging due to varied lighting conditions, minor manufacturing defects, and subsequent pill wear. Surmounting these challenges is possible using multiple inputs: pill color, imprint, and shape. Of these different inputs, pill shape is the most important and difficult parameter due to its variations. In this paper, we describe a novel technique to accurately detect the complex pharmaceutical pill shapes using measurements derived from a superimposed adaptable ring centered automatically on either the shape’s centroid or its bounding box midpoint determined based on the measurements from two other rings, namely the inner ring and the outer ring. It is shown that the measurements from the overlays of the adaptable ring suffice to successfully classify the shapes of the pills currently in the Pillbox database (U.S. National Library of Medicine, 2014) with an accuracy of 98.7%. Our method demonstrated higher accuracy when compared with Hu-moments on the same data set. Using logistic regression techniques, Hu-moments provided an accuracy of 96.6%. Though developed for the domain of pharmaceutical pill shapes, we discuss how the measurements from the adaptable ring can also be used in other industrial applications to increase the level of accuracy with the help of this real-time less computationally complex method. [ABSTRACT FROM AUTHOR]

The 2016 IEEE Educational Activities Board (EAB), chaired by IEEE Educational Activities Vice President S. K. Ramesh, upon the recommendation of the EAB Awards and Recognition Committee (ARC) has named the recipients of the 2016 IEEE EAB Awards. EAB Awards recognize and honor individuals and companies for major contributions to engineering and technical education. Awards are given for meritorious activities in accreditation, continuing education, educational innovation, preuniversity education, service to the IEEE EAB, standards education, employee professional development, and informal education systems, as well as related achievements that advance the practice of engineering and engineering education. [ABSTRACT FROM PUBLISHER]

Conductor (copper) foil surface roughness in printed circuit boards (PCBs) is inevitable due to adhesion with laminate dielectrics. Surface roughness limits data rates and frequency range of application of copper interconnects and affects signal integrity (SI) in high-speed electronic designs. In measurements of dielectric properties of laminate dielectrics using traveling-wave techniques, conductor surface roughness may significantly affect accuracy of measuring dielectric constant (DK) and dissipation factor (DF), especially at frequencies above a few gigahertz, when copper roughness is comparable to skin depth of copper. This paper proposes an algorithm for characterization of copper foil surface roughness. This is done by analyzing the microsection images of copper foils obtained using optical or scanning electron microscopes. The statistics obtained from numerous copper foil roughness profiles allows for introducing a new metric for roughness characterization of PCB interconnects and developing “design curves,” which could be used by SI engineers in their designs. [ABSTRACT FROM AUTHOR]

Multi-modal data fusion techniques are commonly used to enhance decision-making processes. In previous research, a comprehensive structural analysis process was developed for quantizing and evaluating characteristics of defects in aircraft lap-joint mimics using eddy current (EC) nondestructive evaluation (NDE) data collected for structural health monitoring. In this research, a comprehensive multi-modal structural analysis process is presented that includes intra- and inter-modal NDE data fusion based on EC, millimeter wave (MW), and ultrasonic (UT) data obtained from five lap-joint mimic test panels. The process includes defect detection, defect characterization, and finite-element modeling-based simulated fatigue loading for structural analysis. The multi-modal structural analysis process is evaluated using four test panels with corroded patches at different layers of the lap joints and one painted pristine panel used as a reference. The test panels are subjected to two rounds of mechanical loading, preceded by multi-modal NDE data obtained before each round. Different NDE modality combinations are examined for test panel modeling, including: 1) EC, 2) UT, 3) MW, 4) EC and UT, 5) EC and MW, and 6) EC, UT, and MW. Experiments are performed to compare the simulated fatigue loading, based on models determined from the different modality combinations, and the mechanical loading results to find susceptible-to-failure areas in the test panels. Experimental results showed that the EC and UT modality combination yielded a correct vulnerable (crack) location recognition rate of 98.8%, an improvement of 14.7% over any individual modality, demonstrating the potential for multi-modal data fusion for characterizing corrosion and defects. [ABSTRACT FROM PUBLISHER]

A real-world laboratory exercise is presented for image processing and related curricula. The exercise is a traffic-monitoring problem in which a truck must be tracked as it moves across a bridge and its velocity measured. Sequential images are taken from a dedicated Web camera that views the Smart Composite Bridge on the University of Missouri-Rolla campus. The prototype bridge is a field laboratory for several interdisciplinary courses, including a Machine Vision elective. The Machine Vision image-processing elective uses the traffic-monitoring exercise to give students experience with processing complex images, tracking image markers, and applying theoretical orthographic concepts. The laboratory exercise uses an image sequence acquired during the springtime with multiple potential markers available on the truck for assignment flexibility. A wintertime image sequence with snowy conditions is also available for assignment flexibility. This paper discusses the bridge and camera resources, the traffic-monitoring laboratory exercise description, and the Machine Vision course implementation and evaluation. Two versions of the traffic-monitoring exercise, including two image sequences and orthographic MATLAB code, are available on the bridge Website. [ABSTRACT FROM AUTHOR]