EFFICIENT LINE DETECTION METHOD BASED ON 2D CONVOLUTION FILTER

The article proposes an efficient line detection method using a 2D convolution filter. The proposed method was compared with the Hough transform, the most popular method of straight lines detection. The developed method is suitable for local detection of straight lines with a slope from -45˚ to 45˚. Also, it can be used for curve detection which shape is approximated with the short straight sections. The new method is characterized by a constant computational cost regardless of the number of set pixels. The convolution is performed using the logical conjunction and sum operations. Moreover, design of the developed filter and the method of filtration allows for parallelization. Due to constant computation cost, the new method is suitable for implementation in the hardware structure of real-time image processing systems.

Inspection Method of Rope Arrangement in the Ultra-Deep Mine Hoist Based on Optical Projection and Machine Vision

Due to the importance of safety detection of the drum’s rope arrangement in the ultra-deep mine hoist and the current situation whereby the speed, accuracy and robustness of rope routing detection are not up to the requirements, a novel machine-vision-detection method based on the projection of the drum’s edge is designed in this paper. (1) The appropriate position of the point source corresponding to different reels is standardized to obtain better projection images. (2) The corresponding image processing and edge curve detection algorithm are designed according to the characteristics of rope arrangement projection. (3) The Gaussian filtering algorithm is improved to adapt to the situation that the curve contains wavelet peak noise when extracting the eigenvalues of the edge curve. (4) The DBSCAN (density-based spatial clustering of applications with noise) method is used to solve the unsupervised classification problem of eigenvalues of rope arrangement, and the distance threshold is calculated according to the characteristics of this kind of data. Finally, we can judge whether there is a rope arranging fault just through one frame and output the location and number of the fault. The accuracy and robustness of the method are verified both in the laboratory and the ultra-deep mine simulation experimental platform. In addition, the detection speed can reach 300 fps under the premise of stable detection.

Automatic processing of time domain induced polarization data using supervised artificial neural networks

SUMMARY Processing of geophysical data is a time consuming task involving many different steps. One approach for accelerating and automating processing of geophysical data is to look towards machine learning (ML). ML encompasses a wide range of tools, which can be used to automate complicated and/or tedious tasks. We present strategies for automating the processing of time-domain induced polarization (IP) data using ML. An IP data set from Grindsted in Denmark is used to investigate the applicability of neural networks for processing such data. The Grindsted data set consists of eight profiles, with approximately 2000 data curves per profile, on average. Each curve needs to be processed, which, using the manual approach, can take 1–2 hr per profile. Around 20 per cent of the curves were manually processed and used to train and validate an artificial neural network. Once trained, the network could process all curves, in 6–15 s for each profile. The accuracy of the neural network, when considering the manual processing as a reference, is 90.8 per cent. At first, the network could not detect outlier curves, that is where entire chargeability curves were significantly different from their spatial neighbours. Therefore, an outlier curve detection algorithm was developed and implemented to work in tandem with the network. The automatic processing approach developed here, involving the neural network and the outlier curve detection, leads to similar inversion results as the manual processing, with the two significant advantages of reduced processing times and enhanced processing consistency.

When Not to Classify: Anomaly Detection of Attacks (ADA) on DNN Classifiers at Test Time

A significant threat to the recent, wide deployment of machine learning–based systems, including deep neural networks (DNNs), is adversarial learning attacks. The main focus here is on evasion attacks against DNN-based classifiers at test time. While much work has focused on devising attacks that make small perturbations to a test pattern (e.g., an image) that induce a change in the classifier's decision, until recently there has been a relative paucity of work defending against such attacks. Some works robustify the classifier to make correct decisions on perturbed patterns. This is an important objective for some applications and for natural adversary scenarios. However, we analyze the possible digital evasion attack mechanisms and show that in some important cases, when the pattern (image) has been attacked, correctly classifying it has no utility---when the image to be attacked is (even arbitrarily) selected from the attacker's cache and when the sole recipient of the classifier's decision is the attacker. Moreover, in some application domains and scenarios, it is highly actionable to detect the attack irrespective of correctly classifying in the face of it (with classification still performed if no attack is detected). We hypothesize that adversarial perturbations are machine detectable even if they are small. We propose a purely unsupervised anomaly detector (AD) that, unlike previous works, (1) models the joint density of a deep layer using highly suitable null hypothesis density models (matched in particular to the nonnegative support for rectified linear unit (ReLU) layers); (2) exploits multiple DNN layers; and (3) leverages a source and destination class concept, source class uncertainty, the class confusion matrix, and DNN weight information in constructing a novel decision statistic grounded in the Kullback-Leibler divergence. Tested on MNIST and CIFAR image databases under three prominent attack strategies, our approach outperforms previous detection methods, achieving strong receiver operating characteristic area under the curve detection accuracy on two attacks and better accuracy than recently reported for a variety of methods on the strongest (CW) attack. We also evaluate a fully white box attack on our system and demonstrate that our method can be leveraged to strong effect in detecting reverse engineering attacks. Finally, we evaluate other important performance measures such as classification accuracy versus true detection rate and multiple measures versus attack strength.