Feature fusion for object detection at one map

Xi, Xing; Wu, Yuanqing; Xia, Canming; He, Shenghuang

doi:10.1016/j.imavis.2022.104466

Cited by 6 publications

(2 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For the vessel identification in the preprocessed satellite images, the YOLOv4 neural network architecture [60] was selected. The YOLO network is a CNN-based state-of-the-art solution that has proven to be very efficient in similar tasks of determining moving objects in images (ranging from moving cars to aircrafts) [61][62][63][64][65]. Providing high accuracy with low response time, the YOLOv4 is highly suitable for handling large volumes of data in an effective manner.…”

Section: Vessel Detection Based On Satellite Imagesmentioning

confidence: 99%

The Big Picture: An Improved Method for Mapping Shipping Activities

Troupiotis-Kapeliaris,

Zissis,

Bereta

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Density maps support a bird’s eye view of vessel traffic, through providing an overview of vessel behavior, either at a regional or global scale in a given timeframe. However, any inaccuracies in the underlying data, due to sensor noise or other factors, evidently lead to erroneous interpretations and misleading visualizations. In this work, we propose a novel algorithmic framework for generating highly accurate density maps of shipping activities, from incomplete data collected by the Automatic Identification System (AIS). The complete framework involves a number of computational steps for (1) cleaning and filtering AIS data, (2) improving the quality of the input dataset (through trajectory reconstruction and satellite image analysis) and (3) computing and visualizing the subsequent vessel traffic as density maps. The framework describes an end-to-end implementation pipeline for a real world system, capable of addressing several of the underlying issues of AIS datasets. Real-world data are used to demonstrate the effectiveness of our framework. These experiments show that our trajectory reconstruction method results in significant improvements up to 15% and 26% for temporal gaps of 3–6 and 6–24 h, respectively, in comparison to the baseline methodology. Additionally, a use case in European waters highlights our capability of detecting “dark vessels”, i.e., vessel positions not present in the AIS data.

show abstract

Section: Vessel Detection Based On Satellite Imagesmentioning

confidence: 99%

The Big Picture: An Improved Method for Mapping Shipping Activities

Troupiotis-Kapeliaris,

Zissis,

Bereta

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…It can be obtained by making modifications on the basis of AOCN-C. Referring to the modification methods in the image field [33], a multi-channel one-dimensional adaptive function based channelwise is designed as shown in Equation ( 6) to obtain the one-dimensional Meta ACON-C activation function that is used in this paper.…”

Section: Meta Acon-cmentioning

confidence: 99%

Fault Diagnosis Method of Rolling Bearing Based on CBAM_ResNet and ACON Activation Function

Qin

Pan

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

In order to cope with the influences of noise interference and variable load on rolling bearing fault diagnosis in real industrial environments, a rolling bearing fault diagnosis method based on CBAM_ResNet and ACON activation function is proposed. Firstly, the collected bearing working vibration signals are made into input samples to retain the original features to the maximum extent. Secondly, the CBAM_ResNet fault diagnosis model is constructed. By taking advantage of the convolutional neural network (CNN) in classification tasks and key feature extraction, the convolutional block attention module network (CBAM) is embedded in the residual blocks, to avoid model degradation and enhance the interaction of information in channel and spatial, raise the key feature extraction capability of the model. Finally, the Activate or Not (ACON) activation function, is introduced to adaptively activate shallow features for the purpose of improving the model’s feature representation and generalization capability. The bearing dataset of Case Western Reserve University (CWRU) is used for experiments, and the average accuracy of the proposed method is 97.68% and 93.93% under strong noise interference and variable load, respectively. Compared with the other three published bearing fault diagnosis methods, the results indicate that this proposed method has better noise immunity and generalization ability, and has good application value.

show abstract

Weather-degraded image semantic segmentation with multi-task knowledge distillation

Wu²,

Wang³

et al. 2022

Image and Vision Computing

View full text Add to dashboard Cite

Feature fusion for object detection at one map

Cited by 6 publications

References 2 publications

The Big Picture: An Improved Method for Mapping Shipping Activities

The Big Picture: An Improved Method for Mapping Shipping Activities

Fault Diagnosis Method of Rolling Bearing Based on CBAM_ResNet and ACON Activation Function

Weather-degraded image semantic segmentation with multi-task knowledge distillation

Contact Info

Product

Resources

About