Squeeze and Excitation Rank Faster R-CNN for Ship Detection in SAR Images

Zhao, Lin; Ji, Kefeng; Leng, Xiangguang; Kuang, Gangyao

doi:10.1109/lgrs.2018.2882551

Cited by 316 publications

(166 citation statements)

References 14 publications

Supporting

Mentioning

127

Contrasting

Order By: Relevance

“…(1) YOLO divides each input image into S × S (S = 7) grids, each of which is responsible for predicting the targets that fall into the center of the grid, as shown in Figure 3 (3) After the S × S × B bounding boxes are predicted, the confidence values are multiplied by the respective predicted class probability to obtain the class confidence [31,33]. The classification threshold (CT) of the class confidence is set to filter the bounding boxes and the classes whose class confidence is less than that of CT. (4) Next, we set the the Non-maximum suppression algorithm threshold (NT) and use it to filter the redundant frame for the reserved bounding box [15,16,20,21,31,33].…”

Section: Monitoring Principlementioning

confidence: 99%

“…These results indicate problems with current methods for gathering traffic-flow statistics.Today, with the aid of artificial intelligence, deep learning is used for target detection, semantic segmentation, image classification, and other identification tasks in various scenarios. The region-based method is the most common for target detection, while using the R-CNN, SPP-net, Fast R-CNN, and Faster R-CNN algorithms [14][15][16][17][18][19][20][21]. R-CNN uses a selective search to extract regions from an image, which is efficient, but training is cumbersome, and its test speed is slow [14,22].…”

mentioning

confidence: 99%

“…Fast R-CNN combines the R-CNN and SPP-net concepts and it can improve detection speed and accuracy, but the speed is still relatively slow [15,16]. Faster R-CNN uses an RPN network instead of selective search algorithms, which greatly shortens the time to extract regions from images, but only achieves seven frames per second for video detection, falling short of real-time detection requirements [20][21][22]. In order to improve the detection speed and accuracy, based on the DenseNet model [23], the researchers proposed a lightweight PeleeNet model for mobile devices, which had the highest target classification accuracy [24].…”

mentioning

confidence: 99%

See 2 more Smart Citations

Investigation of a Promoted You Only Look Once Algorithm and Its Application in Traffic Flow Monitoring

et al. 2019

View full text Add to dashboard Cite

We propose a high-performance algorithm while using a promoted and modified form of the You Only Look Once (YOLO) model, which is based on the TensorFlow framework, to enhance the real-time monitoring of traffic-flow problems by an intelligent transportation system. Real-time detection and traffic-flow statistics were realized by adjusting the network structure, optimizing the loss function, and introducing weight regularization. This model, which we call YOLO-UA, was initialized based on the weight of a YOLO model pre-trained while using the VOC2007 data set. The UA-CAR data set with complex weather conditions was used for training, and better model parameters were selected through tests and subsequent adjustments. The experimental results showed that, for different weather scenarios, the accuracy of the YOLO-UA was ~22% greater than that of the YOLO model before optimization, and the recall rate increased by about 21%. On both cloudy and sunny days, the accuracy, precision, and recall rate of the YOLO-UA model were more than 94% above the floating rate, which suggested that the precision and recall rate achieved a good balance. When used for video testing, the YOLO-UA model yielded traffic statistics with an accuracy of up to 100%; the time to count the vehicles in each frame was less than 30 ms and it was highly robust in response to changes in scenario and weather.

show abstract

Section: Monitoring Principlementioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

See 1 more Smart Citation

Investigation of a Promoted You Only Look Once Algorithm and Its Application in Traffic Flow Monitoring

et al. 2019

View full text Add to dashboard Cite

show abstract

“…Assuming that the intersection over union (IOU) between the reference box and ground truth (GT) of Anchor was greater than 0.7, it is marked as a positive sample; and when the overlap (IOU) between the reference box and GT of Anchor is less than 0.3, it will be marked as a negative sample. As for the remaining samples that are neither positive nor negative, they will not participate in the final training [19].…”

Section: Loss Functionmentioning

confidence: 99%

A Ship Target Location and Mask Generation Algorithms Base on Mask RCNN

Shaodan¹,

Feng²,

Chen³

2019

IJCIS

View full text Add to dashboard Cite

Ship detection is a canonical problem in computer vision. Motivated by the observation that the major bottleneck of ship detection lies on the different scales of ship instances in images, we focus on improving the detection rate, especially for the smallsized ships which are relatively far from the camera. We use the Smooth function combined with L1 and L2 norm to optimize the region proposal network (RPN) loss function and reduce the deviation between the prediction frame and the actual target to ensure the accurate location of the ship target. With the Two-Way sampling combined with the shared weight to generate the mask, we solve the problems of inaccurate segmentation, target loss and small interference when Mask Region Convolution Neural Network (RCNN) is used to segment an instance. We create the experimental data sets from the deep learning annotation tool-Labelme. Experiments show that the improved Mask-RCNN model has a confidence rate of 82.17%. Serving as the basic network, the test accuracy rate of ResNetXt-101 is 3.3% higher than that of the original ResNet-101, which can better realize the function of ship target location and mask generation.

show abstract

“…The construction of the FPN aims to extract high-resolution and segmentation features by combining the output of the BU and TD pathways, but it takes a long time and consumes memory. At the same time, the development of computational processor devices like the graphics processing units (GPUs) have contributed to the improvement and development of image classification and recognition by introducing effective methods, like the fully convolutional network (FCN) [37], residual network (ResNet) and squeeze and excitation (SENet) [20,21,[37][38][39].…”

Section: Introductionmentioning

confidence: 99%

Multi-Scale Geospatial Object Detection Based on Shallow-Deep Feature Extraction

et al. 2019

View full text Add to dashboard Cite

Multi-class detection in remote sensing images (RSIs) has garnered wide attention and introduced several service applications in many fields, including civil and military fields. However, several reasons make detection from aerial images very challenging and more difficult than nature scene images: Objects do not have a fixed size, often appear at very various scales and sometimes appear in dense groups, like vehicles and storage tanks, and have different surroundings or background areas. Furthermore, all of this makes the manual annotation of objects very complex and costly. The powerful effect of the feature extraction methods on object detection and the successes of deep convolutional neural networks (CNN) extract deep features more than traditional methods. This study introduced a novel network structure and designed a unique feature extraction which employs squeeze and excitation network (SENet) and residual network (ResNet) to obtain feature maps, named a shallow-deep feature extraction (SDFE), that improves the resolution and the localization at the same time. Furthermore, this novel model reduces the loss of dense groups and small objects, and provides higher and more stable detection accuracy which is not significantly affected by changing the value of the threshold of the intersection over union (IoU) and overcomes the difficulties of RSIs. Moreover, this study introduced strong evidence about the factors that affect the detection of RSIs. The proposed shallow-deep and multi-scale (SD-MS) method outperforms other approaches for the given ten classes of the NWPU VHR-10 dataset.

show abstract

Squeeze and Excitation Rank Faster R-CNN for Ship Detection in SAR Images

Cited by 316 publications

References 14 publications

Investigation of a Promoted You Only Look Once Algorithm and Its Application in Traffic Flow Monitoring

Investigation of a Promoted You Only Look Once Algorithm and Its Application in Traffic Flow Monitoring

A Ship Target Location and Mask Generation Algorithms Base on Mask RCNN

Multi-Scale Geospatial Object Detection Based on Shallow-Deep Feature Extraction

Contact Info

Product

Resources

About