Because clouds and snow block the underlying surface interfere with information extracted from an image, accurate segmentation of cloud/snow regions is essential for imagery preprocessing remote sensing. Nearly all sensing images have a high resolution contain complex diverse content, which makes task more difficult. A multi-branch convolutional attention network (MCANet) suggested in this study. double-branch structure adopted, spatial semantic image are extracted. In way, model’s feature...

During the preceding biennium, vision-language pre-training has achieved noteworthy success on several downstream tasks. Nevertheless, acquiring high-quality image-text pairs, where pairs are entirely exclusive of each other, remains a challenging task, and noise exists in commonly used datasets. To address this issue, we propose SoftCLIP, novel approach that relaxes strict one-to-one constraint achieves soft cross-modal alignment by introducing softened target, which is generated from...

Distracted driver actions can be dangerous and cause severe accidents. Thus, it is important to detect eliminate distracted driving behaviors on the road save lives. To this end, we study action detection using videos captured inside vehicle. We propose Stargazer, an efficient, transformer-based system exploiting rich temporal features about human behavioral information, with a simple yet effective localization framework. The core of our contains improved version multi-scale vision...

The analysis of land cover types is helpful for detecting changes in use categories and evaluating resources. It great significance environmental monitoring, management, planning, mapping. At present, remote sensing imagery obtained by widely employed the classification types. However, most existing methods have problems such as low accuracy, vulnerability to noise interference, poor generalization ability. Here, a multi-scale contextual semantic guidance network proposed deep learning....

The Greater Mekong Subregion (GMS) economic cooperation program is an effective and fruitful regional initiative for socioeconomic development in Asia; however, the vegetation change trends directions GMS caused by rapid remain unknown. In particular, there a current lack of comparative studies on changes various countries GMS. Based MODIS normalized difference index (NDVI) time series data, this study analyzed spatiotemporal patterns coverage their from 2000 to 2022 using Theil–Sen slope...

Video surveillance has become more and prevalent. It is a basic problem to get the number of access people in scenes. When occlusions occur, it becomes difficult count people. We propose fast robust counting method, implement system. In our system, we use group tracking compensate weakness multiple human segmentation, which can handle complete occlusion. Our system run real-time about 30fps for CIF video, with accuracy defined by frame above 95%.

Current training objectives of existing person Re-IDentification (ReID) models only ensure that the loss model decreases on selected batch, with no regards to performance samples outside batch. It will inevitably cause over-fit data in dominant position (e.g., head imbalanced class, easy or noisy samples). The latest resampling methods address issue by designing specific criterion select trains generalize more certain type hard samples, tail data), which is not adaptive inconsistent real...

Abstract Electrically conductive cellulose-based hydrogels are prepared by a facile and environmentally friendly method, of which the electrical mechanical properties can be easily controlled varying graphene loading. With an ultralow initial addition oxide (GO, 0.2 wt% versus mass cellulose), resulting cellulose/reduced (CG ) hydrogel shows significantly enhanced compressive modulus 332.01 kPa, 54.8% higher than that pure cellulose hydrogel. Further increasing GO to 2 (versus conductivity...

With the rapid development of hardware equipments, it is now economically and technically feasible to build a video surveillance system. This paper presents system architecture VISS, intelligent deployed in parking lots. In VISS we adopt robust moving object detecting tracking algorithm, present novel activity recognition framework based on layer hidden semi-Markov model (LHSMM) which used for modeling activities. The experimental results real-time shows effective complex

Cloud detection is an important prerequisite for remote sensing image application. Any from which the information of ground object could be obtained will inevitably preprocessed on cloud occlusion. In traditional method, segmentation and its shadow affected by complex background. process, due to insufficient extraction, misjudgment often occurs, boundary processing also very rough. order improve accuracy segmentation, we propose a multilevel feature context semantic fusion network. The...

The classification of land cover types is an important task for monitoring use. Moreover, with the continuous application high resolution remote sensing images, time and space span becoming larger larger, which greatly increases difficulty target types. And there are few results that can effectively deal deep category information. Furthermore, extraction fusion features still need to be improved. In this paper, a multi-angle attention network (MAFNet) proposed uses 50-layer residual as...

Image Quality Assessment (IQA) with reference images have achieved great success by imitating the human vision system, in which image quality is effectively assessed comparing query its pristine image. However, for wild, it quite difficult to access accurate images. We argue that possible learn knowledge under No-Reference (NR-IQA) setting, effective and efficient empirically. Concretely, innovatively introducing a novel feature distillation method IQA, we propose new framework comparative...

Background subtraction is a typical method to segment foreground from background in motion analysis. Most of methods base on static backgrounds. We propose particle filter based estimate dynamic backgrounds, such as water waves, foliage, smoke, etc. consider the scenes vary under some rules and use track background. Experiments show our results are promising, compared with other

Obstacle avoiding is one of the most complex tasks for autonomous driving systems, which was also ignored by many cutting-edge end-to-end learning-based methods. The difficulties stem from integrated process detection and interpretation environment obstacles generation proper behaviors. We make use CARLA, a simulator research, collect massive human drivers' reactions to on road subjecting given commands, i.e. follow, go straight, turn left right about 6 hours. A behavior-Cloning neural...

Image Quality Assessment (IQA) models benefit significantly from semantic information, which allows them to treat different types of objects distinctly. Currently, leveraging information enhance IQA is a crucial research direction. Traditional methods, hindered by lack sufficiently annotated data, have employed the CLIP image-text pretraining model as their backbone gain awareness. However, generalist nature these pre-trained Vision-Language (VL) often renders suboptimal for IQA-specific...

Abstract Geological hazards seriously affect the personal and property safety of regional residents, in Xichuan area, a real-time monitoring early warning system could have provided timely alerts before landslide, potentially averting tragic loss life destruction.So it is very important to carry out geological hazard susceptibility assessment prevention. The Xixia area selected as study support vector machine model used evaluate area. Through Kendall factor correlation analysis, evaluation...

Regional geohazard susceptibility evaluation and early warning are effective means of disaster prevention mitigation. The traditional regional has problems such as limited model accuracy insufficient refinement. With the rapid development big data artificial intelligence technology, machine learning algorithms gradually widely used in geologic hazard have achieved better results. paper uses BP neural network support vector to predict susceptibility. selects Utopia District Shiyan City, Hubei...

Real-time multi-target path planning is a key issue in the field of autonomous driving. Although multiple paths can be generated real-time with polynomial curves, are not flexible enough to deal complex road scenes such as S-shaped and unstructured parking lots. Search sampling-based methods, A* RRT their derived generating for these environments. However, existing algorithms require significant time plan targets, which greatly limits application In this paper, method multi-targets proposed....