IJIGSP Vol. 14, No. 6, Dec. 2022
Cover page and Table of Contents: PDF (size: 660KB)
Currently, the means of semantic segmentation of images, which are based on the use of neural networks, are increasingly being used in computer systems for various purposes. Despite significant progress in this industry, one of the most important unsolved problems is the task of adapting a neural network model to the conditions for selecting an object mask in an image. The features of such a task necessitate determining the type and parameters of convolutional neural networks underlying the encoder and decoder. As a result of the research, an appropriate method has been developed that allows adapting the neural network encoder and decoder to the following conditions of the segmentation problem: image size, number of color channels, acceptable minimum segmentation accuracy, acceptable maximum computational complexity of segmentation, the need to label segments, the need to select several segments, the need to select deformed , displaced and rotated objects, allowable maximum computational complexity of training a neural network model, allowable training time for a neural network model. The main stages of the method are related to the following procedures: determination of the list of image parameters to be registered; formation of training example parameters for the neural network model used for object selection; determination of the type of CNN encoder and decoder that are most effective under the conditions of the given task; formation of a representative educational sample; substantiation of the parameters that should be used to assess the accuracy of selection; calculation of the values of the design parameters of the CNN of the specified type for the encoder and decoder; assessment of the accuracy of selection and, if necessary, refinement of the architecture of the neural network model. The developed method was verified experimentally on examples of semantic segmentation of images containing objects such as a car. The obtained experimental results show that the application of the proposed method allows, avoiding complex long-term experiments, to build a NN that, with a sufficiently short training period, ensures the achievement of image segmentation accuracy of about 0.8, which corresponds to the best systems of similar purpose. It is shown that it is advisable to correlate the ways of further research with the development of approaches to the use of special modules such as ResNet, Inception and mechanisms of the Partial convolution type used in modern types of deep neural networks to increase their computational efficiency in the encoder and decoder.[...] Read more.
In this paper, an efficient technique for anomalous pedestrian activity detection in the academic institution is proposed. At the pixel and block levels, the proposed method elicits motion components that accurately represent pedestrian action, velocity, and direction, as well as along a frame. We also adopted these motion features to detect anomalous actions. The detection of anomalous behavior in academic environments is not available at the moment. Similarly, the existing method produces a high number of false positives. An anomaly detection dataset and a newly designed proposed student behavior database were used to validate the proposed framework. A significant improvement in anomalous activity recognition has been demonstrated in experimental results. Based on motion features, the proposed method reduces false positives by 3% and increases true positives by 5%. A discussion of future research directions concludes the paper.[...] Read more.
The aviation security is an important component of aviation safety providing. One of the main goals of aviation security service is to detect dangerous and prohibited objects during passengers and baggage screening. For this purpose, aviation security personnel use various equipment: X-ray screening system, body-scans, metal detectors, moving ions detectors, explosive trace detectors. The X-ray screening system gives information on internal structure of baggage. The main disadvantage of X-ray screening system is rather high level of the false alarm probability. This requires developing new methods of image processing and recognition of dangerous and prohibited objects on the background of other objects. This article develops the principles of shadow image processing while screening the baggage using X-ray system to fix the mentioned disadvantage. The math equation for shadow image is obtained based on the laws of geometry and Beer-Lambert equation taking into account the chosen scanning technique. Based on this, the article is focused to the analysis of simple objects images and their application for complex objects recognition. The article discusses the example of handgun recognition using a new approach based on spectral analysis of developed shadow images. The results of the research can be used for improvement of algorithmic toolkit in aviation security automatic decision-making system while screening the baggage by X-ray equipment.[...] Read more.
Today, rain remains one key and well-known natural phenomenon that offsets and attenuates the propagated radio, microwave, and millimeter-wave signals at different transmission frequencies and wavelengths over propagation paths. Specialised rain attenuation studies can be utilized to analyze their stochastic behavior on propagated radio signals and also come up with appropriate rain attenuation model for network application planning and optimisations. In this contribution, empirical rainfall depths data has been acquired, effectively categorized, and employed to examine the implicative intensity level trends over a ten years period, starting from 2011 to 2020. More importantly, the Recommendation ITU-R P.1511 power-based model combined with the acquired categorized rainfall depths data has been explored to prognostically estimate and quantity the amount of specific attenuation loss due over 3.5G transmission frequency. The results reveal that the level of attenuation attained versus 0.01% percentage of time depends on the type of rain intensity levels (heavy rain, very heavy rain, extremely heavy rain), which in turn is dependent upon rain depth or rate drop sizes. As a case in point, 0.001 percent of the time due to heavy rain, the amount of specific attenuation attained stood at 2dB, while for very heavy and extremely heavy rain, the specific attenuation levels amount to 2.3dB and 4dB respectively. These different amounts of specific attenuation simplify imply that the heavier the rain, the more scattering, and absorption the propagated electromagnetic signals undergo, thus leading to degraded and higher attenuation levels. The empirical based-rain attenuation quantification and impact analysis method explored in this paper will significantly provide radio network engineers with the best way to monitor and evaluate the radio attenuation effect over a propagation channel.[...] Read more.
This paper analyses the performance of various lossless compression algorithms employed on uniformly quantized audio signals. The purpose of this study is to enlighten a new way of audio signal compression using lossless compression algorithms. The audio signal is first transformed into text by employing uniform quantization with different step sizes. This text is then compressed using lossless compression algorithms which include Run length encoding (RLE), Huffman coding, Arithmetic coding and Lempel-Ziv-Welch (LZW) coding. The performance of various lossless compression algorithms is analyzed based on mainly four parameters, viz., compression ratio, signal-to-noise ratio (SNR), compression time and decompression time. The analysis of the aforementioned parameters has been carried out after uniformly quantizing the audio files using different step sizes. The study exhibits that the LZW coding can be a potential alternative to the MP3 lossy audio compression algorithm to compress audio signals effectively.[...] Read more.
Deep learning (DL) architectures are becoming increasingly popular in modern traffic systems and self-driven vehicles owing to their high efficiency and accuracy. Emerging technological advancements and the availability of large databases have made a favorable impact on such improvements. In this study, we present a traffic sign recognition system based on novel DL architectures, trained and tested on a locally collected traffic sign database. Our approach includes two stages; traffic sign identification from live video feed, and classification of each sign. The sign identification model was implemented with YOLO architecture and the classification model was implemented with Xception architecture. The input video feed for these models were collected using dashboard camera recordings. The classification model has been trained with the German Traffic Sign Recognition Benchmark dataset as well for comparison. Final accuracy of classification for the local dataset was 96.05% while the standard dataset has given an accuracy of 92.11%. The final model is a combination of the detection and classification algorithms and it is able to successfully detect and classify traffic signs from an input video feed within an average detection time of 4.5fps[...] Read more.
This paper proposes hierarchical support vector machines for weapon Identification using images of repeated stab wound patterns caused by sharp metal weapons used in homicidal cases and also presents a comparative study with standard flat support vector machine. The methodology includes the segmentation technique for the extraction of region-of-interest in the image using transition region-based segmentation algorithm and then texture, shape and size features were extracted from the segmented image. For multiple classes, a hierarchical support vector machine is adapted as a classifier. This approach gives a computationally interesting and efficient alternative solution to identify the weapons used in the crime; this method uses the digital images of repeated stab wound patterns which appear on the human body. The experimental study has three main stages, at the first stage includes generating of non-overlapping segments from the pattern, at the second stage the features of wound patterns are extracted and finally identification of patterns and its weapon of cause. The proposed method accuracy assessment is performed and also comparison study is performed with standard flat support vector machine and with the traditional method of forensic pathology. The experimental results achieved for Identification is 96.71% accuracy, with an available database of 500 images of a pattern consisting of repeated stabbed wounds. From the comparative study, the proposed methodology has given better results than standard SVM and traditional method. The proposed method delivers a better solution for identification from the image of the repeated stab wound pattern as there is no human intervention that reduces the error and data manipulation unlike traditional manual method.[...] Read more.