Automated vehicle detection in satellite images using deep learning

Mansour, Ahmad; Hassan, Ahmed S.; Hussein, Walid; Said, Ehab

doi:10.1088/1757-899x/610/1/012027

Cited by 21 publications

(8 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In general, our model's mAP is similar to models trained on aerial images from other research. The Multi-Scale CNN by Guo et al [43] achieves a mAP of 89.6%, the Faster R-CNN and SSD models by Mansour et al [45] achieve 89.2% and 84.2% and the YOLO based model by Haroon et al [44] achieves 60.9%. However, most objects detected in these studies are significantly larger and therefore easier to detect for a CNN than many small ACC condensers in our study [43,44,45].…”

Section: Detection Of Air-cooled Condensers and Cooling Towersmentioning

confidence: 99%

“…Typically, a computer model learns how to process given data in order to make predictions for unseen data [41]. This technique is already frequently used to efficiently solve time consuming tasks such as detecting water wells, storage tanks, various vehicles, planes, boats, and certain buildings within satellite and aerial images [42,43,44,45,46,47,48]. In addition, previous studies detected CTs to support the investigation of Legionnaires' disease outbreaks in the US [49].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Estimating cooling capacities from aerial images using convolutional neural networks

Barth¹,

Schüppler²,

Menberg³

et al. 2023

Preprint

View full text Add to dashboard Cite

In recent decades, the global cooling demand has significantly increased and is expected to grow even further in the future. However, knowledge regarding the spatial distribution of cooling demand is sparse. Most existing studies are based on statistical modelling, which lack in small-scale details and cannot accurately identify individual large cooling producers. In this study, we implement and apply a novel method to identify, map and estimate nominal cooling capacities of chillers using deep learning. Chillers typically use air-cooled condensers and cooling towers to release excess heat, and produce most of the cooling needs in the commercial and industrial sectors. In this study, these units are identified from aerial images using specifically trained object detection models. The corresponding nominal cooling capacity is then estimated based on the number of fans of air-cooled condensers and the fan diameters of the cooling towers, respectively. Both detection and capacity estimations are first evaluated on test data sets and subsequently applied to an industrial area (Brühl) and the city center in Freiburg, Germany. In Brühl, aerial images show chillers with an estimated nominal cooling capacity of 201 MW, of which the model detected 88%, while 92% of all detections are correct. In the city center, a nominal capacity of 18.6 MW is estimated, of which the model detected 87% with 77% of all detections being correct. Hence, the developed approach facilitates a reliable analysis of the installed nominal cooling capacity of individual buildings at large scales, such as districts and cities. This information could be further used to locate areas for investments and support planning of eco-friendly, centralized supply of cooling energy, for example district heating and cooling systems or shallow geothermal energy systems such as aquifer thermal energy storage (ATES).

show abstract

Section: Detection Of Air-cooled Condensers and Cooling Towersmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Estimating cooling capacities from aerial images using convolutional neural networks

Barth¹,

Schüppler²,

Menberg³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Ferdous et al [12] introduced prediction speed criteria in 2019 and argued that Regions-CNN (RCNN) [17], Fast-RCNN [16], and Faster-RCNN [47] are incompatible for real-time applications due to slow multistage regional-proposal based approach. End-to-end detection-based methods like You Only Look Once (YOLO) [46] and Single Shot Detectors (SSD) [34], where suggested to increase prediction speed, yet compromising on accuracy (only 89.21%) [38]. An alternative fully convolutional neural network was proposed by Shelhamer et al [51] that combined features from complementary resolution levels (contextual and spatial information).…”

Section: Related Workmentioning

confidence: 99%

Deep learning-based object recognition in multispectral satellite imagery for real-time applications

Gudzius

Kurasova

Darulis

et al. 2021

Machine Vision and Applications

View full text Add to dashboard Cite

Satellite imagery is changing the way we understand and predict economic activity in the world. Advancements in satellite hardware and low-cost rocket launches have enabled near-real-time, high-resolution images covering the entire Earth. It is too labour-intensive, time-consuming and expensive for human annotators to analyse petabytes of satellite imagery manually. Current computer vision research exploring this problem still lack accuracy and prediction speed, both significantly important metrics for latency-sensitive automatized industrial applications. Here we address both of these challenges by proposing a set of improvements to the object recognition model design, training and complexity regularisation, applicable to a range of neural networks. Furthermore, we propose a fully convolutional neural network (FCN) architecture optimised for accurate and accelerated object recognition in multispectral satellite imagery. We show that our FCN exceeds human-level performance with state-of-the-art 97.67% accuracy over multiple sensors, it is able to generalize across dispersed scenery and outperforms other proposed methods to date. Its computationally light architecture delivers a fivefold improvement in training time and a rapid prediction, essential to real-time applications. To illustrate practical model effectiveness, we analyse it in algorithmic trading environment. Additionally, we publish a proprietary annotated satellite imagery dataset for further development in this research field. Our findings can be readily implemented for other real-time applications too.

show abstract

“…The performances of object detection methods were compared using deep learning models such as Vgg16, ResNet-101, Inception-Resnet-v2 (Jiao et al, 2019). In a study conducted for vehicle detection, Vgg-16 and Inception-V2 were used and the success rates were found to be over 80% (Mansour et al, 2019). There is no sample comparing the performance of deep learning models according to the application results in the studies conducted.…”

Section: Related Workmentioning

confidence: 99%

“…In a 5-class object detection study using images with 30cm and 4.8m width, it was observed that the mAP value was 0.53 in images with a width of 30cm, while this performance decreased to 0.11 in images with 4.8m (Shermeyer&Van Etten, 2019). In a study for vehicle detection, Vgg-16 and Inception-V2 were used and the success rates were found to be over 80% (Mansour et al, 2019). There is no sample comparing the performance of deep learning models according to the results of the application in the studies.…”

Section: Related Workmentioning

confidence: 99%

Comparison of deep learning models in terms of multiple object detection on satellite images

Doğan

Türkoğlu²

2021

JER is an international, peer-reviewed journal that publishes f

View full text Add to dashboard Cite

The images obtained by remote sensing contain important data about ground surface. It is an important issue to detect objects on the ground surface with these images. Deep learning models are known to give better results in studies on object detection. However, the superiority of the deep learning models over each other is unknown. For this reason, it should be clarified which model is superior in terms of object detection and which model should be used in studies. In this study, it was aimed to reveal the superiorities of deep learning models by comparing their performance in detecting multiple objects. By using 11 deep learning models that are frequently encountered in the literature, the application of detecting objects of 14 classes in the DOTA dataset were made. 49,053 objects in 888 images were used for training by using AlexNet, Vgg16, Vgg19, GoogleNet, SequezeeNet, Resnet18, Resnet50, Resnet101, Inceptionresnetv2, inceptionv3, DenseNet201 models. After the training, 13,772 objects consisting of 14 classes in 277 images were used for testing with RCNN, which is one of the object detection methods. The performance of each algorithm in 14 classes has been demonstrated by using Average Precision (AP) and Mean Average Precision (mAP) to measure the performance of the models from their metrics. In a particular class of each deep learning model, difference in performance was observed The model with the highest performance varies in each class. In the application, the most successful average mAP value of 14 classes was Vgg16 with 24.64, while the lowest was InceptionResnetV2 with 11.78. In this article, the success of deep learning models in detecting multiple objects has been demonstrated practically and it is thought to be an important resource for researchers who will study on this subject.

show abstract

Automated vehicle detection in satellite images using deep learning

Cited by 21 publications

References 13 publications

Estimating cooling capacities from aerial images using convolutional neural networks

Estimating cooling capacities from aerial images using convolutional neural networks

Deep learning-based object recognition in multispectral satellite imagery for real-time applications

Comparison of deep learning models in terms of multiple object detection on satellite images

Contact Info

Product

Resources

About