Storm-Drain and Manhole Detection Using the RetinaNet Method

Santos, Anderson Aparecido dos; Marcato, José; Silva, Jonathan de Andrade; Pereira, Rodrigo Bahia; Matos, Daniel Anijar de; Menezes, Geazy Vilharva; Higa, Leandro Tsuneki; Eltner, Anette; Ramos, Ana Paula Marques; Osco, Lucas Prado; Gonçalves, Wesley Nunes

doi:10.3390/s20164450

Cited by 27 publications

(12 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…RetinaNet combines the advantages of multiple target recognition methods, especially the “anchor” concept introduced by Region Proposal Network (RPN) [ 14 ], and the use of feature pyramids in Single Shot Multibox Detector (SSD) [ 20 ] and Feature Pyramid Networks (FPN) [ 21 ]. Retinanet has a wide range of applications, such as ship detection in remote sensing images of different resolutions [ 22 ], identification of storm drains and manholes in urban areas [ 23 ], fly identification [ 24 ], and rail surface crack detection [ 25 ]. The experiment described in this paper was conducted on a computer equipped with Intel ® Xen(R) W-2145 CPU and NVIDIA GeForceRTX 2080Ti.…”

Section: Methodsmentioning

confidence: 99%

Wheat Ear Recognition Based on RetinaNet and Transfer Learning

Fei

et al. 2021

Sensors

View full text Add to dashboard Cite

The number of wheat ears is an essential indicator for wheat production and yield estimation, but accurately obtaining wheat ears requires expensive manual cost and labor time. Meanwhile, the characteristics of wheat ears provide less information, and the color is consistent with the background, which can be challenging to obtain the number of wheat ears required. In this paper, the performance of Faster regions with convolutional neural networks (Faster R-CNN) and RetinaNet to predict the number of wheat ears for wheat at different growth stages under different conditions is investigated. The results show that using the Global WHEAT dataset for recognition, the RetinaNet method, and the Faster R-CNN method achieve an average accuracy of 0.82 and 0.72, with the RetinaNet method obtaining the highest recognition accuracy. Secondly, using the collected image data for recognition, the R2 of RetinaNet and Faster R-CNN after transfer learning is 0.9722 and 0.8702, respectively, indicating that the recognition accuracy of the RetinaNet method is higher on different data sets. We also tested wheat ears at both the filling and maturity stages; our proposed method has proven to be very robust (the R2 is above 90). This study provides technical support and a reference for automatic wheat ear recognition and yield estimation.

show abstract

Section: Methodsmentioning

confidence: 99%

Wheat Ear Recognition Based on RetinaNet and Transfer Learning

Fei

et al. 2021

Sensors

View full text Add to dashboard Cite

show abstract

“…In our implementation, we found that d = 0.3 m and C min = 10,000 works well to remove the majority of these ghost points without removing other relevant clusters. The ground segmentation is performed by applying a cloth simulation filter [26]. This filter inverts the point cloud along the Z-direction and simulates a cloth being dropped on top of it.…”

Section: Sor CC Csf and Roi Filteringmentioning

confidence: 99%

“…The classification threshold is the minimum distance to classify a point as ground or nonground. The optimal parameter settings are based on the findings in [26] and a few experimental tests. We chose 1, 3 and 0.3 m for the grid resolution, rigidity and classification threshold, respectively.…”

Section: Sor CC Csf and Roi Filteringmentioning

confidence: 99%

Manhole Cover Detection on Rasterized Mobile Mapping Point Cloud Data Using Transfer Learned Fully Convolutional Neural Networks

Mattheuwsen

Vergauwen

2020

Remote Sensing

View full text Add to dashboard Cite

Large-scale spatial databases contain information of different objects in the public domain and are of great importance for many stakeholders. These data are not only used to inventory the different assets of the public domain but also for project planning, construction design, and to create prediction models for disaster management or transportation. The use of mobile mapping systems instead of traditional surveying techniques for the data acquisition of these datasets is growing. However, while some objects can be (semi)automatically extracted, the mapping of manhole covers is still primarily done manually. In this work, we present a fully automatic manhole cover detection method to extract and accurately determine the position of manhole covers from mobile mapping point cloud data. Our method rasterizes the point cloud data into ground images with three channels: intensity value, minimum height and height variance. These images are processed by a transfer learned fully convolutional neural network to generate the spatial classification map. This map is then fed to a simplified class activation mapping (CAM) location algorithm to predict the center position of each manhole cover. The work assesses the influence of different backbone architectures (AlexNet, VGG-16, Inception-v3 and ResNet-101) and that of the geometric information channels in the ground image when commonly only the intensity channel is used. Our experiments show that the most consistent architecture is VGG-16, achieving a recall, precision and F2-score of 0.973, 0.973 and 0.973, respectively, in terms of detection performance. In terms of location performance, our approach achieves a horizontal 95% confidence interval of 16.5 cm using the VGG-16 architecture.

show abstract

“…First Faster R-CNN detects utility poles and the second one crop on the top half of these detected objects to detect the cap pole missing. Moreover, RetinaNet outperformed other deep learning methods in several remote sensing applications [15,20,21]. Despite these initial efforts, there is a lack of research focusing on utility poles automatic mapping using aerial images, mainly using novel deep learning methods, such as Adaptive Training Sample Selection (ATSS).…”

Section: Introductionmentioning

confidence: 99%

“…We compared the achieved results with Faster R-CNN [23] and RetinaNet [24], which are common methods applied in remote sensing image analysis. RetinaNet, for example, outperformed other deep learning methods in several remote sensing applications [15,20,21].…”

Section: Introductionmentioning

confidence: 99%

Mapping Utility Poles in Aerial Orthoimages Using ATSS Deep Learning Method

Gomes

Silva

Gonçalves

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

Mapping utility poles using side-view images acquired with car-mounted cameras is a time-consuming task, mainly in larger areas due to the need for street-by-street surveying. Aerial images cover larger areas and can be feasible alternatives although the detection and mapping of the utility poles in urban environments using top-view images is challenging. Thus, we propose the use of Adaptive Training Sample Selection (ATSS) for detecting utility poles in urban areas since it is a novel method and has not yet investigated in remote sensing applications. Here, we compared ATSS with Faster Region-based Convolutional Neural Networks (Faster R-CNN) and Focal Loss for Dense Object Detection (RetinaNet ), currently used in remote sensing applications, to assess the performance of the proposed methodology. We used 99,473 patches of 256 × 256 pixels with ground sample distance (GSD) of 10 cm. The patches were divided into training, validation and test datasets in approximate proportions of 60%, 20% and 20%, respectively. As the utility pole labels are point coordinates and the object detection methods require a bounding box, we assessed the influence of the bounding box size on the ATSS method by varying the dimensions from 30×30 to 70×70 pixels. For the proposal task, our findings show that ATSS is, on average, 5% more accurate than Faster R-CNN and RetinaNet. For a bounding box size of 40×40, we achieved Average Precision with intersection over union of 50% (AP50) of 0.913 for ATSS, 0.875 for Faster R-CNN and 0.874 for RetinaNet. Regarding the influence of the bounding box size on ATSS, our results indicate that the AP50 is about 6.5% higher for 60×60 compared to 30×30. For AP75, this margin reaches 23.1% in favor of the 60×60 bounding box size. In terms of computational costs, all the methods tested remain at the same level, with an average processing time around of 0.048 s per patch. Our findings show that ATSS outperforms other methodologies and is suitable for developing operation tools that can automatically detect and map utility poles.

show abstract

Storm-Drain and Manhole Detection Using the RetinaNet Method

Cited by 27 publications

References 36 publications

Wheat Ear Recognition Based on RetinaNet and Transfer Learning

Wheat Ear Recognition Based on RetinaNet and Transfer Learning

Manhole Cover Detection on Rasterized Mobile Mapping Point Cloud Data Using Transfer Learned Fully Convolutional Neural Networks

Mapping Utility Poles in Aerial Orthoimages Using ATSS Deep Learning Method

Contact Info

Product

Resources

About