مجله پردازش سیگنال پیشرفته
سال چهارم شماره 2 (پیاپی 6، پاییز و زمستان 1399)

Recently, small object recognition based on deep learning techniques has gained particular attention in many practical applications and is challenging because small objects have low resolution and do not contain detailed information.  In this article, a new two-stage detector based on detecting objects with recursive feature pyramid and switchable atrous convolution (DetectoRS) has been introduced to find small and important defects such as loose nut-bolts and missing-nut in power transmission lines (PTL). The architecture of DetectoRS was necessarily modified. The proposed technique which is called DRSPTL, the Cascade R-CNN with ResNext-101 is used to increase the accuracy of small defect detection. In this work, high-resolution RGB images are captured by unmanned aerial vehicles (UAVs) imaging PTL from Tehran, Kerman, Shiraz, Isfahan, and Ahwaz regional electric companies, Iran. The training and test datasets from the captured faulty images are created from annotation by experts. To construct the training dataset, nearly eighty percent of the whole set of faulty images were selected and labeled. The performance of the proposed method with two state-of-the-art object detection techniques RetinaNet and RepPoints has been compared. DRSPTL has the highest small defect detection accuracy. It is noteworthy that the obtained results could significantly reduce the time and cost of electric power companies by detecting the defects automatically and preventing the occurrence of many power outages.

The importance of understanding three-dimensional spacer fabrics properties, is a reason to the development of rapid and accurate methods for determining properties, due to their numerous applications in various industries. In most applications, spacer fabrics are affected by tension; therefore, knowing their behaviour in the encounter with tension is important. The purpose of this paper is to investigate on deformation of the spacer fabric and determine the local displacements in this fabric under tension. So, the digital image correlation method was used, that is a usual method of determining the displacements and deformation of a structure under external loading. Deformation behaviour of the diamond shape unit the of spacer fabrics structure at different tensile strains, based on experimental observations and theoretical analysis using video processing and digital image correlation method was investigated in the course and wale direction. The fabric unit deformation, the distribution of the local displacements and longitudinal and transverse strain of fabric were determined using video processing and compared with experimental method. Comparison of results showed that video processing method is able to calculate the local displacement in fabric and predict the longitudinal and transverse strain at different tensile strains with an error coefficient less than 10%.

The huge number and volume of video and video usage have caused most of them saved and transferred as compressed video. Nowadays, indexing, searching and retrieving video directly in compressed domain has been taken great attention. The first step in video indexing and retrieval is segmenting a video into chronological sets and manageable pieces or shots. In recent years, a new video coding standard say H.265 has been introduced and it is needed to develop the methods and algorithms for analyzing, indexing and retrieving H.265 compressed video. In this paper, a novel method has been proposed for shot detection in H.265 compressed video without full decompression. In the proposed method, macro block coding information which is in headers of compressed H.265 bit-stream by a threshold is used for shot detection of compressed video. The results indicate that the proposed method detect video shots in H.265 video by 80% accuracy and 84/3% recall.

Encryption is one of the most powerful tools that ensures information security in the field of communication and information technology. Image encryption is different from other encryptions. This difference is due to the inherent characteristics of the images. Recent attempts to encrypt images have been based on chaos. In this paper, a new chaos-based encryption algorithm for image encryption is presented. In the proposed method, instead of encryption one image in each step, four images are encrypted simultaneously. In this way, four standard images are combined and a single image of their combination is created. Simultaneous encoding of four images will cause to complicate the proposed encryption algorithm, increase security, and extend the amount of gray area per pixel. Finally, the resulting image is XORed with encryption key and the encrypted image is generated. Considering the combination of four images and their simultaneous encoding, as well as examining a large number of image evaluation criteria, specifically the information entropy, that has achieved a value equal to 7/9994 in our proposed algorithm, which is really close to the ideal value of 8, shows that the proposed algorithm performs appropriately.

Facial image segmentation is an essential component in applications of image processing and computer vision such as face recognition, identity recognition and analysis of facial plastic surgery. The clustering based methods are one of the important methods in the facial image segmentation. Self-Organizing Map (SOM) is a powerful method in the data mining. A main disadvantage of the SOM algorithm is that learning coefficient is not adaptive in this algorithm. Adaptability of learning coefficient in the adaptation phase can improve the performance of the SOM clustering. Neural Gas Network (NGN) is an unsupervised learning that its neighborhood structure is adaptive and synaptic weight is updated without any topological adjustment. The main purpose of this study is to present a new hybrid SOMNGN method in which the learning coefficient is adapted in the adaptation phase of the SOM algorithm using the NGN algorithm. Also, two color spaces, including YCbCr and Face Mapping are used for facial skin modelling as a pre-processing step. Obtained results in the mentioned color spaces show that presented method have the higher accuracy than the standard SOM method.

Diabetic Retinopathy (DR) is one of the major complications of Diabetes, which is the injury to the retina of the diabetic patient and causes blindness if not diagnosed in early stages. Various machine learning classification and clustering approaches have been studied in literature with the purpose of improving the accuracy of the screening methods. Some of machine learning classification and clustering approaches are based on manually feature extraction of fundus images by image processing experts. In recent years, a new approach for image classification and diagnosis without using any manual feature extraction is proposed based on convolutional neural network (CNN). In medical imaging and diagnosis, training a deep CNN from scratch is difficult because it requires a large amount of labeled training data and the training procedure is a time consuming task to ensure proper convergence. Therefore, a very common method to train CNNs for medical diagnosis is fine-tuning a pre-trained CNN. In this paper, the pre-trained GoogleNet as a powerful CNN is employed on the Kaggle database for DR diagnosis from retinal images. To assess the efficacy of the clinical results, the proposed CNN algorithm is performed to diagnose DR from the images that are gathered from the the Navid-Didegan ophthalmology clinic.

Improving the performance of relay systems using patterns such as power allocation requires access to instantaneous channel coefficients . extraction of this information is always prone to errors and this error increases the undesirable portions of the received signal in the receiver , in addition to disrupting the relay system optimization pattern . the Fixed and variable Gain relays , according to the type of operation , have different effect on the estimation error of channel coefficients . in this paper , with a detailed examination of the effect of the estimation error of channel coefficients according to the pattern of estimation , suitable power allocation is proposed to reduce the effects of undesirable segments in the Fixed and variable Gain operation. Reference papers in this matter for power allocation always use second - order statistical data but in this paper , the estimated coefficients will be used . due to the complexity of the performance of the variable gain relay in control of the interference channel effect in two-way relay , the closed form for power allocation is presented.

The problem of automatic handwritten context recognition has received considerable attention of many researchers. In this paper, a fusion system is proposed to enhance the recognition accuracy of Farsi handwritten digits. The proposed approach consists of a prepration process and two main phases. In the prepration process, some pre-processing operations are performed on the image. Then some features are extracted, among which a multi-objective particle swarm optimization selects more effective ones. For every image, these optimal features are given as the input data to the classifiers. In the first main phase, training datasets are used to construct three different SVMs. In order to achieve better results, the adaptive best-mass gravitational search algorithm is utilized to adjust the SVMs parameters. In the second main phase, an interval type–II fuzzy inference system receives the SVMs outputs and by combining them, it presents a more accurate estimation of the digit in the image. The results of applying the proposed approach to the problem of scanned Farsi handwritten digits in the standard HODA database demonstrated that this algorithm attains high accuracy, precision and recall performance indices, comparing to other existing methods.

In this paper, a lossy compression method with the ability to control the quality of the reconstructed signal is proposed for phonocardiography (PCG) signals. It is based on two main ideas: down-sampling and two-dimensionalization. For PCG image compression, wavelet transform and Spatial-oriented Tree Wavelet (STW) encoder are used. In the proposed method, there is the ability to control the quality of the reconstructed signal using a Percent Root-mean-square Difference (PRD)-related threshold. The simulation results of the proposed method on some public databases indicates that the down-sampling stage has a significant effect on increasing the compression ratio especially in the case of databases with high sampling frequency. The next important factor in improving the compression efficiency of the proposed method is the use of two-dimensional PCG signal in order to take advantage of the inter-period redundancy in this type of repetitive signals, and using modern effective methods for image compression. The efficiency of the proposed method was evaluated according to the average PRD and Compression Ratio (CR) criteria and compared with the results of several existing methods. In this evaluation, while limiting PRD≤5%, the lowest average compression ratio was related to the Artifacts dataset from the Pascal database (with a sampling frequency of 2000 Hz) and the highest average compression ratio was related to the database of the University of Washington (with a sampling frequency of 44100 Hz).

Recognition of silent speech only by decoding brain signals is one of the latest research in the field of artificial intelligence. The 3rd Iranian national brain-computer interface competition, which was held by the National Brain Mapping Center of Iran in 2020, was dedicated to the classification of imagined speech for the three words of rock-paper-scissors game. In this contest, the authors introduced an approach based on wavelet packets decomposition and common spatial pattern and could win the second place. We evaluated some of the most famous classifiers including support vector machine, k-nearest neighbor, random forest, logistic regression, XGBoost and dense deep learning model separately for each subject and simultaneously for all subjects. The best average accuracy was 51.7%. Then we developed the model using spectrogram and convolutional neural network and achieved average accuracy of 76.5%. The accuracy was much better than the accuracy reported by other researchers on this dataset. Also, the performance of our model is superior to recent research in this field on different datasets.

This paper presents an optimized multi-digit Ternary to Binary converter based on nano-carbone tubes field-effect transistors. By modifying a part of the circuit structure of the ternary-to-binary converter, the efficiency of the system has increased. Due to the unique features nanotubes carbon tubes feild effect transistors, as well as the possibility of designing different threshold voltages for transistors, designing multi-level logic systems is much simpler and less costly. Therefore, considering that the existing processing systems work on a dual basis, the design of binary to bernary converters and vice versa is very important and basic processing systems. Therefore, considering that the existing processing systems work on a binary, the design of binary to turner and turner to binary converters is very important and fundamental in processing systems. The circuit modification has reduced chip occupancy, reduced power consumption, and reduced circuit latency. The proper and optimal performance of the proposed converter have been confirmed by simulation by HSPICE software based on 32 nm CNTFET transistor. The simulation results show that the optimal terbnary to binary converter has a power consumption of 0.665 μW and a propagation delay of 27.3 ps. These results show that overall PDP index has improved by 14.4%.

Today, the use of automated optical inspection systems in the production of printed circuit boards to control solders, the presence of the right elements and their direction has become an essential tool for electronic companies. The printed circuit board in this system is irradiated by several light sources and one or more high-definition cameras are used for imaging. Automated optical inspection system, using the recorded image and comparing the image information with the machine information, detects and specifies any type of error (defect) or suspicious areas. In this paper, using a camera mounted on a conveyor, we try to cover most of the common errors that occur on printed circuit boards at any stage of the production line. The traveling salesman algorithm is used to control the movement of the camera on the conveyor. To introduce the printed circuit board to the system, a software has been designed that uses a CAD file to obtain the location and type of elements on the board. By selecting the optimal camera movement path, it detects errors due to the absence of elements, direction of elements, lack of soldering, cold soldering, excessive soldering, etc. in three stages of feature extraction, feature selection and decision making. The results show that the device is efficient in detecting glue error before installing the elements and detecting errors after tin bath