AUTOMATIC CONTROL AND COMPUTER SCIENCES最新文献

In response to the characteristics of financial image data, this paper proposes an efficient digital image compression scheme. Firstly, discrete cosine transform (DCT) is applied to divide the financial image into DC and AC coefficients. Secondly, based on the characteristics of DCT coefficients, a fuzzy method is employed to categorize DCT subblocks into smooth, texture, and edge classes, enabling distinct quantization strategies. Subsequently, to eliminate spatial and statistical redundancies in financial images, common features and structures are utilized, and a specific scanning approach is employed to optimize the arrangement of important coefficients. Finally, differential prediction and entropy coding are employed for DCT coefficient scanning encoding, enhancing compression efficiency. The objective evaluation metrics of this algorithm are approximately 2 dB higher than existing algorithms at bit rates of 0.25 and 0.5. Even at bit rates of 0.75, 1.5, 2.5, and 3.5, the performance of this method still outperforms the comparative algorithms, demonstrating its capability to efficiently store and transmit massive financial image data, thereby providing robust support for data processing in the financial sector.

Arabic handwritten character recognition (AHCR) is the process of automatically identifying and recognizing handwritten Arabic characters. This is a challenging task due to the complexity of the Arabic script, which includes a large number of characters with complex shapes and ligatures. In this paper, we present a novel approach based on Levenshtein distance to recognize Arabic handwritten characters by combining the classification and the postprocessing phases. To train the proposed model, we created an Arabic optical character recognition (OCR) context database divided into multiple text files. Each file in the database belongs to one of five well-defined contexts: sport, economy, religion, politics, and culture. The total number of words in each file is 15 000. The experiment results show that the new method outperforms the state-of-the-art approach. The error rate achieved by using 15 000 words was 1.2%.

In pursuit of enhancing public safety and addressing challenges in driver behavior recognition, an intelligent recognition and detection method of driver behavior based on ResNet (IRDMDB-ResNet) is proposed. The approach aims to identify instances of distracted driving resulting from abnormal behavior. Three models (IRDMDB-1, IRDMDB-2, and IRDMDB-3) are presented to implement this method, which is adapted to a deep learning behavior recognition in driving scenarios. Firstly, this study utilizes two well-tested real datasets: Driver Drowsiness Dataset and The State Farm. These datasets undergo preprocessing to meet the input requirements of the model. Secondly, a lightweight convolutional neural network model has been designed to extract features, aiding the warning system in delivering precise information and minimizing traffic collisions to the maximum extent possible. Finally, the model is evaluated based on the confusion metrics, accuracy, precision, recall, and F1-score criterion. As a result, the IRDMDB-3 model proposed in this paper can recognize and detect driver behavior effectively and stably. And it achieves 99.79% of accuracy in the classification of distracted drivers looking elsewhere in The State Farm dataset. Similarly, the detection at Driver Drowsiness Dataset is 99.68%. This advancement represents a significant improvement in traffic safety, showcasing adaptability to diverse behaviors and remarkable recognition and detection capabilities.

For the elderly, falls can be extremely fatal. However, due to the physical decline of the elderly, it is difficult to avoid falls. Therefore, to the greatest extent feasible lessen the harm that falls on the elderly inflict, so that they can be found in the first time of falls, this study based on wearable devices, proposed a fall monitoring system using an improved K-nearest neighbor algorithm. The improved fuzzy K-nearest neighbor algorithm combined with support vector machine algorithm is applied to improve the efficiency and accuracy of the algorithm, and reduce the false positive rate and false negative rate as much as possible. The suggested model’s average precision in the simulation experiment is 97.5%. The specificity was 97.6%. The sensitivity was 97.5%. The convergence performance is also good, 24 iterations can reach the optimal. In the actual experiment, the average accuracy reached 98.7%; The false alarm rate is only 0.7%; The negative rate was 2.5%; Its performance is superior to other two algorithms. This shows that the proposed method has excellent accuracy, false positive rate and false negative rate in practical application, which has important significance for the health and safety of the elderly.

This paper presents a new method of localizing radio frequency (RF) source in non-line of sight (NLOS) using data collected using the anchor and map. The measurable observation in the unmanned aerial vehicle (UAV) is assumed to be the received signal strength indicator (RSSI), and a method is presented based on the RSSI observation of the reflected signal sent from the anchor to estimate the location of the reflecting obstacle, which is a two-step method for map estimation and localization. It is also assumed that the map of the obstacle location is also available; the location of the reflective obstacle can be obtained using the map with an error. And finally, by combining this data in a weighted and improved particle filter for the optimal use of the number of particles in a wide area, the location of the unknown RF source is estimated more accurately. It was revealed that the proposed method improved localization and had good precision.

In this study, the Tamarix chinensis forest in Changyi national marine ecological special protected area in Shandong province, China, was researched for forest fire monitoring based on thermal infrared remote sensing technology. We summarized the commonly monitoring methods for forest fire point based on remote sensing technology into two types: fixed threshold method (including its deformation model and extension model) and adjacent pixel analysis method (also known as background pixel correlation method). And we analyzed the advantages and disadvantages of these two methods. The BT (brightness temperature) data inverted from the remote sensing images of IRS sensor (HJ 1B satellite) and TIRS sensor (Landsat-8 satellite) indicated that there not had enough thermal radiation to form a fire point during the above phases in the protected zone. The research results and methods also confirmed that thermal infrared remote sensing technology can be used for forest fire monitoring and identification of macro forest fire point.

In this paper, on the basis of in-depth research on the key technology of binocular vision measurement; a set of multidimension online measurement system for image recognition is built. Canny operator is used as a tool to detect the contour features of parts, and the Canny operator is accelerated and improved from the aspects of mathematical reasoning and Gaussian pyramid. A synchronous external trigger circuit for a binocular camera and light source was designed. Finally, the improved algorithms in various aspects of visual measurement in this paper are applied to the measurement system. The experimental results show that the online measurement system has the advantages of high measurement accuracy and small repeatability errors.

Facial features extracted from deep convolutional networks are susceptible to background, individual identity and other factors. It interferes with facial expression recognition when mixed with useless features. Considering that different scale features have rich semantic and texture information respectively, this paper takes VGG-16 as the basic network structure and combines multiscale features to obtain richer feature information. In addition, the input feature map elements are enhanced or suppressed by the attention module in order to extract salient features more accurately. The proposed method was validated on two commonly used expression data sets CK+ and RAF-DB, and the recognition rates were 98.77 and 82.83%, respectively. Experimental results show the superiority of this method.

This paper deals with the problem of creating a keyword spotting (KWS) system with real-world audio data. The paper describes the different methods used to build KWS systems, deep learning models such as convolutional neural networks (CNN), transformers, etc. The paper also discusses the mainstream dataset for training and testing KWS models, Google Speech Commands. We conduct experiments on Google Speech Commands dataset and propose our method of creating a KWS dataset and that helps neural networks achieve better results in training on relatively small amounts of data. We also introduce an idea of a hybrid KWS inference system architecture that uses voice detection and light-weight speech recognition framework in attempt to boost its computational performance and accuracy. We conclude by noting that KWS is an important challenge in the field of speech recognition, and suggest that their method can be used to improve the performance of KWS systems in the circumstances of low amounts of training data. We also note that future research could focus on bettering the process of evaluating the models and improving the overall performance of KWS systems.

In recent years, the proliferating of IoT (Internet of things)-originated applications have generated huge amounts of data, which has put enormous pressure on infrastructures such as the network cloud. In this regard, scholars have proposed an architectural model for “cloud-fog” computing, where one of the obstacles to fog computing is how to allocate computing resources to minimize network resources. A heuristic-based TDCC (Time, distance, cost and computing-power) algorithm is proposed to optimize the task scheduling problem in this heterogeneous system for genetic algorithm-based “cloud-fog” computing, including execution time, operational cost, distance and total computing power resources. The algorithm uses evolutionary genetic algorithms as a research tool to combine the advantages of cloud computing, fog computing and genetic algorithms to achieve a balance between latency, cost, link length and computing power. In the hybrid computing task scheduling, this algorithm has a better balance than TCaS algorithm which only considers a single metric; this algorithm has a better adaptation value than traditional MPSO algorithm by 2.61%, BLA algorithm by 6.92% and RR algorithm by 33.39%, respectively. The algorithm is also flexible enough to match the user’s needs for high performance distance-cost-computing power, enhancing the effectiveness of the system.