Training set effect on super resolution for automated target recognition

Ciolino, Matthew; Noever, David; Kalin, Josh

doi:10.1117/12.2557845

Cited by 3 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this study, the datasets consisting of images taken from different environmental conditions are used to train the network to obtain a high-quality SRR model [64]. Adding images containing the geometric characteristics of rock cracks in the dataset is conducive to improving the reconstruction effect of the SRR network for micro-cracks [65]. Dataset I and Dataset III are the input to the SRR network for training.…”

Section: Performance Comparison Of Srr Modelsmentioning

confidence: 99%

Development and application of a more refined process for extracting rock crack width information based on artificial intelligence

Ye¹,

Yu²,

Zhang

et al. 2023

Preprint

View full text Add to dashboard Cite

The process of image collection of high-altitude rock cracks using unmanned aerial vehicle (UAV) suffers from insufficient resolution and motion blur, which prevents more accurate detection of micro-cracks. Therefore, in this study, a rock crack refinement detection process (RC-RDP) based on super-resolution reconstruction (SRR) technique and semantic segmentation (SS) network is developed to detect micro-cracks. Four SRR networks (RCAN, SRDenseNet, ESRGAN, BSRGAN) and six SS networks (PSPNet, SegNet, DeepLab V3+, UNet++, UNet++(CBAM), SegFormer) are trained and tested separately using rock crack datasets. SRR indicators (PSNR and SSIM) and SS indicators (Precision, Recall, F1-Score and IoU) are applied to evaluate the performance of SRR networks and SS networks, respectively. According to the evaluation indicators of each network performance, in this paper, the RCAN network (PSNR = 31.08 dB and SSIM = 88.56%) is applied in the SRR part, and the UNet++ (CBAM) network is used in the crack detection part (Precision = 0.874, Recall = 0.886, F1-Score = 0.879 and IoU = 0.785). In addition, a novel data acquisition process integrating skeletonization algorithm, feature nodes detection algorithm, normal vector estimation algorithm and width extraction algorithm is used to calculate the minimum width WMin, maximum width WMax and average width WA information of the crack traces. The results of this paper show that the application of RC-RDP based on deep learning can obtain better super resolution (SR) images and more accurate width information. The method in this study shows excellent performance in detecting rock cracks.

show abstract

Section: Performance Comparison Of Srr Modelsmentioning

confidence: 99%

Development and application of a more refined process for extracting rock crack width information based on artificial intelligence

Ye¹,

Yu²,

Zhang

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Regardless, compression ratio and PSNR are used to rank NN compression models. Comparisons between bitrate, the average number of bits needed to encode each image pixel information [16], and performance is a heavily researched topic with Cheng et al [17] [18] [19] discussing its effect across many different types of NNs including Generative Adversarial Networks (GAN) [20], Super Resolution (SR) [21] [22], and others [23].…”

Section: A Backgroundmentioning

confidence: 99%

Image Compression and Actionable Intelligence With Deep Neural Networks

Ciolino¹

2022

Preprint

Self Cite

View full text Add to dashboard Cite

If a unit cannot receive intelligence from a source due to external factors, we consider them disadvantaged users. We categorize this as a preoccupied unit working on a low connectivity device on the edge. This case requires that we use a different approach to deliver intelligence, particularly satellite imagery information, than normally employed. To address this, we propose a survey of information reduction techniques to deliver the information from a satellite image in a smaller package. We investigate four techniques to aid in the reduction of delivered information: traditional image compression, neural network image compression, object detection image cutout, and image to caption. Each of these mechanisms have their benefits and tradeoffs when considered for a disadvantaged user.

show abstract

“…This appears particularly important as the training can have a dramatic impact on performance. 11,13 Various methods have been proposed, specifically to synthesize visible-like images from thermal images and vice-versa. These include works to synthesize RGB (visible) images from thermal images for face recognition applications 14 or driving scenes.…”

Section: Introductionmentioning

confidence: 99%

Perspectives and limitations of visible-thermal image pair synthesis via generative adversarial networks

Panchard

Marelli

Presa³

et al. 2021

Target and Background Signatures VII

View full text Add to dashboard Cite

Many applications rely on thermal imagers to complement or replace visible light sensors in difficult imaging conditions. Recent advances in machine learning have opened the possibility of analyzing or enhancing images, yet these methods require large annotated databases. Training approaches that leverage data augmentation via simulated and synthetically-generated images could offer promising prospects. Here, we report on a method that uses generative adversarial nets (GANs) to synthesize images of a complementary contrast. Starting from a dual-modality dataset of co-registered visible and thermal images, we trained a GAN to generate synthetic thermal images from visible images and vice versa. Our results show that the procedure yields sharp synthesized images that might be used to augment dual-modality datasets or assist in visual interpretation, yet are also subject to the limitations imposed by contrast independence between thermal and visible images.

show abstract

Training set effect on super resolution for automated target recognition

Cited by 3 publications

References 0 publications

Development and application of a more refined process for extracting rock crack width information based on artificial intelligence

Development and application of a more refined process for extracting rock crack width information based on artificial intelligence

Image Compression and Actionable Intelligence With Deep Neural Networks

Perspectives and limitations of visible-thermal image pair synthesis via generative adversarial networks

Contact Info

Product

Resources

About