Search

Scene text recognition (STR) has been a hot research field in computer vision, aiming to recognize text in natural scenes using computers. Currently, attention-based encoder–decoder frameworks struggle to precisely align feature regions with the target object when dealing with complex and low-quality images, a phenomenon known as attention drift. Additionally, with the rise of Transformer, the increasing size of parameters results in higher computational costs. In order to solve the above problems, based on the latest research results of Vision Transformer (ViT), we utilize an additional position-enhancement branch to alleviate attention drift and dynamically fused position information with visual information to achieve better recognition accuracy. The experimental results demonstrate that our model achieves a 3% higher average recognition accuracy on the test set compared to the baseline. Meanwhile, our model maintains the advantage of a small number of parameters and fast inference speed, achieving a good balance between accuracy, speed, and computational load.

VQA (Visual Question Answering) is a multi-model task. Given a picture and a question related to the image, it will determine the correct answer. The attention mechanism has become a de facto component of almost all VQA models. Most recent VQA approaches use dot-product to calculate the intra-modality and inter-modality attention between visual and language features. In this paper, the BAN (Bilinear Attention Network) method was used to calculate attention. We propose a deep multimodality bilinear attention network (DMBA-NET) framework with two basic attention units (BAN-GA and BAN-SA) to construct inter-modality and intra-modality relations. The two basic attention units are the core of the whole network framework and can be cascaded in depth. In addition, we encode the question based on the dynamic word vector of BERT(Bidirectional Encoder Representations from Transformers), then use self-attention to process the question features further. Then we sum them with the features obtained by BAN-GA and BAN-SA before the final classification. Without using the Visual Genome datasets for augmentation, the accuracy of our model reaches 70.85% on the test-std dataset of VQA 2.0.

This paper reports a web-based remote collaborative medical diagnosis system. This system provides a dialogue platform using text and speech, by Chinese and Uyghur, containing information such as disease names and symptoms for different medical classifications. The interactive text and speech between doctors or patients can be translated bilingually between Chinese and Uyghur automatically. At last, diagnosis report will be generated for the doctor and the patient. Finally, the performance of the e-health service system is evaluated.

Obstacle-avoiding is a key and challenging issue for autonomous mobile robot navigation, especially in unstructured environments. This paper proposes a fuzzy logic information fusion algorithm based on multi-ultrasonic ranging sensors. Compared to the single sensor approach, this algorithm can not only get the obstacle's distance and orientation information more accurately and reliably, but also obtain the numerical output results of the speed and the turning angle for robot's obstacle avoiding behavior. In addition, the method can simplify the system design by eliminating the necessity of temperature compensation which is routinely included in ultrasonic ranging. MATLAB based simulations and the local path finding experiments on the prototype autonomous mobile robot showed that this algorithm can help the robot timely and robustly to avoid the obstacles in an unstructured environment in a more correct and reliable way, in the same time, the guided path is more smooth and the fluctuation is less and more reasonable. Obviously, the algorithm proposed in the paper is effective.

In recent years, there have been a significant number of activities in the area of Uyghur speech processing. This paper summarized the major tasks and key technologies in Uyghur speech processing, which including the speech database, continuous speech recognition and speech synthesis. Uyghur language is one of the least studied languages on speech processing area. For this reason, in our work, the first step was collecting large amount continuous speech data of Uyghur language. In Uyghur continuous speech recognition, we were building the HMM state models for each recognition unit, and were using the recognizer of HTK3.3 (HMM ToolKit) and the MS Visual C++8.0 developing the basic Uyghur Continuous Speech Recognition System. In Uyghur speech synthesis, we were designing and developing an intelligible and natural sounding corpus-based speech synthesis system.