Pandrol track fastener defect detection based on local convolutional neural networks

The Pandrol track fastener image is composed of two parts: track fastener clip sub-graph and track fastener bolt sub-graph. However, the detection of track fastener clip defect can be realized by track fastener image and track fastener image cannot effectively detect whether the bolt is loose. When the convolutional neural network is used to extract whole picture features and detect, many bolt features unrelated to the clips will be obtained, thereby resulting in a high false alarm rate. To solve these problems, a method based on local convolutional neural network to detect the Pandrol track fastener defects is proposed. First, the algorithm for automatic segmentation of track fastener pictures was used to divide the picture of the Pandrol track fastener into two sub-pictures, one sub-picture is the track fastener bolt and the other sub-picture is the track fastener clip. Second, convolutional neural network was used to detect the track fastener clip pictures. The influence of bolt features unrelated to clips on clips detection can be avoided through image segmentation for local feature extraction, thereby reducing the false alarm rate. Finally, the validity of the proposed method is verified using real Pandrol track fastener images.

Volatility of Tornadogenesis: An Ensemble of Simulated Nontornadic and Tornadic Supercells in VORTEX2 Environments

Despite an increased understanding of the environments that favor tornado formation, a high false-alarm rate for tornado warnings still exists, suggesting that tornado formation could be a volatile process that is largely internal to each storm. To assess this, an ensemble of 30 supercell simulations was constructed based on small variations to the nontornadic and tornadic environmental profiles composited from the second Verification of the Origins of Rotation in Tornadoes Experiment (VORTEX2). All simulations produce distinct supercells despite occurring in similar environments. Both the tornadic and nontornadic ensemble members possess ample subtornadic surface vertical vorticity; the determinative factor is whether this vorticity can be converged and stretched by the low-level updraft. Each of the 15 members in the tornadic VORTEX2 ensemble produces a long-track, intense tornado. Although there are notable differences in the precipitation and near-surface buoyancy fields, each storm features strong dynamic lifting of surface air with vertical vorticity. This lifting is due to a steady low-level mesocyclone, which is linked to the ingestion of predominately streamwise environmental vorticity. In contrast, each nontornadic VORTEX2 simulation features a supercell with a disorganized low-level mesocyclone, due to crosswise vorticity in the lowest few hundred meters in the nontornadic environment. This generally leads to insufficient dynamic lifting and stretching to accomplish tornadogenesis. Even so, 40% of the nontornadic VORTEX2 ensemble members become weakly tornadic. This implies that chaotic within-storm details can still play a role and, occasionally, lead to marginally tornadic vortices in suboptimal storms.

An Automatic Target Recognition Algorithm Based on Support Vector Machine

To improve the performance of automatic target recognition technology and solve the problems of traditional methods, such as high false alarm rate and poor adaptability to environment changes, a new algorithm based on support vector machine is proposed. We have realized the feature extraction of the target and the parameter optimization of the support vector machine to get the support vector machine model applied to the target recognition of unknown images. Experiment results show that the algorithm has a good recognition effect, a fast recognition speed and certain anti-interference abilities based on sufficient samples training.

Correlating data from different sensors to increase the positive predictive value of alarms: an empiric assessment

Objectives: Alarm fatigue from high false alarm rate is a well described phenomenon in the intensive care unit (ICU). Progress to further reduce false alarms must employ a new strategy. Highly sensitive alarms invariably have a very high false alarm rate. Clinically useful alarms have a high Positive-Predictive Value. Our goal is to demonstrate one approach to suppressing false alarms using an algorithm that correlates information across sensors and replicates the ways that human evaluators discriminate artifact from real signal.Methods: After obtaining IRB approval and waiver of informed consent, a set of definitions, (hypovolemia, left ventricular shock, tamponade, hemodynamically significant ventricular tachycardia, and hemodynamically significant supraventricular tachycardia), were installed in the monitors in a 10 bed cardiothoracic ICU and evaluated over an 85 day study period. The logic of the algorithms was intended to replicate the logic of practitioners, and correlated information across sensors in a way similar to that used by practitioners. The performance of the alarms was evaluated via a daily interview with the ICU attending and review of the tracings recorded over the previous 24 hours in the monitor. True alarms and false alarms were identified by an expert clinician, and the performance of the algorithms evaluated using the standard definitions of sensitivity, specificity, positive predictive value, and negative predictive value.Results: Between 1 and 221 instances of defined events occurred over the duration of the study, and the positive predictive value of the definitions varied between 4.1% and 84%.Conclusions: Correlation of information across alarms can suppress artifact, increase the positive predictive value of alarms, and can employ more sophisticated definitions of alarm events than present single-sensor based systems.

Co-ordinated Follow-Up of Transiting Planet Candidates with Robotic Telescope Facilities

There are now several large photometric surveys scanning millions of stellar light-curves for signs of planetary transits. All produce large candidate lists with a high false alarm rate, so that further observations are required to confirm new detections. One such survey, SuperWASP, produced ~150 candidates during the 2007–2008 season. Here we describe our campaign to follow-up 86 of these candidates using the robotic facilities of Las Cumbres Observatory Global Telescope Network and the Tenagra-II robotic telescope in Arizona. The aim of these observations was to eliminate false positives as far as possible ahead of spectroscopic follow-up and to provide additional photometry to help characterise the surviving targets.

The Effects of the Contrast Polarity of Dot-Pair Partners on the Detection of Bilateral Symmetry

Detection of vertical bilateral symmetry has previously been studied in patterns composed of black or white dots on a grey background under four conditions: (a) same contrast (black or white) for all dots (called BB or WW, for ‘all black or all white’); (b) half of the dots black and half white with positive correspondence between symmetrical dot pairs (called MA for ‘matched’); (c) half of the dots black and half white with negative correspondence between symmetrical dot pairs (called OPP for ‘opposite’); and (d) black (white) dots on one side of the axis and white (black) dots on the other (called BW for ‘one side black the other white’). It was found that performance was ordered BB (or WW) = MA > OPP =BW, where > indicates better performance. That experiment was repeated here in experiment 1 with symmetry axes not only at vertical but also at horizontal and the two diagonals. It was found overall that BB = MA > OPP, BW. However, OPP > BW when random trials were included in the analysis but when they were excluded BW > OPP. This was due to a very high false-alarm rate in condition BW which could be accounted for if grouping by colour occurs prior to symmetry detection. In experiment 2 it was shown that vertical-symmetry salience over other orientations remained about the same as OPP patterns progressively changed into BB patterns by varying the percentage same polarity between 0% and 100% in 12%–13% steps. Thus, dot-pair polarity affects performance without affecting relative axis salience, as was also found recently when dot pattern outlines were masked. All of the data indicate that although opposite dot polarity does reduce performance slightly, the symmetry-detection mechanism is remarkably resilient to such perturbation. The high false-alarm rate in the BW condition of experiment 1 may be accounted for by extremely salient global grouping of dots by luminance which effectively creates an integral stimulus which is perceptually difficult to break down into its component dot pairs, prohibiting the required point-by-point matching necessary to reject symmetry detection. The small detrimental effect of nonmatched polarity might be due to the polarity differences masking the grouping of dots into ‘clumps’ on either side of the axis, a process for which there is a great deal of independent evidence.

Vertical Reversal and Lateral Reversal as Components of Inversion in Recognition Memory for Upside-down Faces

Subjects viewed upright photographs of faces, then attempted to recognize them unchanged, vertically reversed or inverted. In two of four conditions, hit scores were lower for inverted than vertically reversed faces, suggesting that lateral reversal is a meaningful component of inversion. The effect was not sufficiently strong, however, to overcome a generally high false-alarm rate for upside-down faces.

Effects of Practice and Instructional Set on the Measurement of Lingual Vibrotactile Thresholds

The effects on lingual vibrotactile thresholds of three different instructional sets and three different practice conditions were determined for 30 normal adult subjects. Results showed no measurable differences between thresholds obtained with the use of the three different instructional sets, but a high false alarm rate occurred for all conditions. When a subject was given practice at obtaining thresholds with a particular instructional set as a prerequisite to threshold data collection, false Mama responses disappeared. Lower (more sensitive) thresholds also were achieved when the practice condition used required the subject to provide three thresholds within 1µ of each other before commencing with actual testing.