remotesensingimage registration technology has important significance in the field of imageprocessing. The registration technology of SAR images has been a hot research in this field, and it is also a challenge. Ai...
详细信息
We propose an embarrassingly simple but very effective scheme for high-quality dense stereo reconstruction: (i) generate an approximate reconstruction with your favourite stereo matcher; (ii) rewarp the input images w...
详细信息
ISBN:
(数字)9781728193601
ISBN:
(纸本)9781728193618
We propose an embarrassingly simple but very effective scheme for high-quality dense stereo reconstruction: (i) generate an approximate reconstruction with your favourite stereo matcher; (ii) rewarp the input images with that approximate model; (iii) with the initial reconstruction and the warped images as input, train a deep network to enhance the reconstruction by regressing a residual correction; and (iv) if desired, iterate the refinement with the new, improved reconstruction. The strategy to only learn the residual greatly simplifies the learning problem. A standard Unet without bells and whistles is enough to reconstruct even small surface details, like dormers and roof substructures in satellite images. We also investigate residual reconstruction with less information and find that even a single image is enough to greatly improve an approximate reconstruction. Our full model reduces the mean absolute error of state-of-the-art stereo reconstruction systems by >50%, both in our target domain of satellite stereo and on stereo pairs from the ETH3D benchmark.
Feature extraction methods of multi-spectral remotesensingimages is of great significance for remotesensingimage analysis, but it still faces some challenges. The ability of traditional feature extraction methods ...
详细信息
The satellite images acquired from long distances are affected by different atmospheric disturbances such as noise and the image quality is degraded. The images thus require pre-processing to preserve the image qualit...
详细信息
As oil palm has become one of the most rapidly expanding tropical crops in the world, detecting and counting oil palms have received considerable attention. Although deep learning has been widely applied to remote sen...
详细信息
ISBN:
(数字)9781728193601
ISBN:
(纸本)9781728193618
As oil palm has become one of the most rapidly expanding tropical crops in the world, detecting and counting oil palms have received considerable attention. Although deep learning has been widely applied to remotesensingimageprocessing including tree crown detection, the large size and the variety of the data make it extremely difficult for cross-regional and large-scale scenarios. In this paper, we propose a cross-regional oil palm tree detection (CROPTD) method. CROPTD contains a local domain discriminator and a global domain discriminator, both of which are generated by adversarial learning. Additionally, since the local alignment does not take full advantages of its transferability information, we improve the local module with the local attention mechanism, taking more attention on more transferable regions. We evaluate our CROPTD on two large- scale high-resolution satellite images located in Peninsular Malaysia. CROPTD improves the detection accuracy by 8.69% in terms of average F1-score compared with the Baseline method (Faster R-CNN) and performs 4.99-2.21% better than other two state-of-the-art domain adaptive object detection approaches. Experimental results demonstrate the great potential of our CROPTD for large-scale, cross-regional oil palm tree detection, guaranteeing a high detection accuracy as well as saving the manual annotation efforts. Our training and validation dataset are available on https://***/rs-dl/CROPTD.
Artificial neural networks (ANNs) have been recognized as universal estimators and widely used in a range of fields starting from computer science, medical, neuroscience, engineering, remotesensing, artificial intell...
详细信息
Artificial neural networks (ANNs) have been recognized as universal estimators and widely used in a range of fields starting from computer science, medical, neuroscience, engineering, remotesensing, artificial intelligence to business and stock markets. In this study the effectiveness of ANNs for recognition of character pattern as an example has been studied. A programming code has been developed to build ANN model system that utilized feedforward backpropagation methodology in learning session and subsequently recognized several predefined alphabet characters. Based on computational resources A 7×5 matrix bit map was achieved for a number of characters such as A, B, C, D, E and F. Backpropagation training was used to adjust the weights of the branches connecting the neuron layers. Thirty-five digitized values (0, 1) against each character were fed to the model as input variables and fetched to thirty-five input neurons of the as-modeled ANN system. The output was considered as specified codes for each character such as 01010, 01010, 01100, 01101, 01110 and 01111 respectively. The training continued till the predefined tolerance limit reached to less than 0.0001. Forward run of the ANN processing was observed capable enough to recognize some irregular pattern in the character. The results showed that with some modifications in the bit pattern the ANN model system could recognize the pattern of the exact character with maximum 37.5 % deformation. Such a model can be exploited to further advancement in patternrecognition system developments.
Hyperspectral images are useful in a variety of fields such as remotesensing, medical diagnosis, and agriculture. But it requires very expensive professional equipment and a lot of time to obtain. In this paper, we p...
详细信息
ISBN:
(数字)9789881476883
ISBN:
(纸本)9781728181301
Hyperspectral images are useful in a variety of fields such as remotesensing, medical diagnosis, and agriculture. But it requires very expensive professional equipment and a lot of time to obtain. In this paper, we propose a deep learning architecture that reconstructs hyperspectral images from RGB images that are easy to acquire in real time. Hyperspectral reconstruction is inherently difficult because much information is lost when hyperspectral bands are integrated into three RGB channels. To effectively overcome the problem of hyperspectral restoration, we design a neural network with the following three basic principles. First, it adopts a method in which channels are gradually increased through several steps to restore hyperspectral images. Second, it is learned on a group basis for efficient restoration. Hyperspectral bands are divided into three groups: R, G, and B. Finally, the concept of channel back projection is newly proposed. In the process of gradually performing hyperspectral reconstruction, the reconstructed image is refined by repeatedly projecting the reconstructed hyperspectral to RGB. In the experimental results, these three principles proved the performance that exceeds the state-of-the-art methods.
Because of the uncertainty of remotesensingimage and ill-posedness for model, the traditional unsupervised classification algorithm is difficult to model accurately in the classification process. The pattern recogni...
详细信息
Current research focuses on non-contact means to capture physiological signals like the heart rate. One promising approach uses videos (imaging PPG, iPPG). The common procedure to derive the heart rate by iPPG compris...
详细信息
ISBN:
(数字)9781728193601
ISBN:
(纸本)9781728193618
Current research focuses on non-contact means to capture physiological signals like the heart rate. One promising approach uses videos (imaging PPG, iPPG). The common procedure to derive the heart rate by iPPG comprises three steps: segmentation of a region of interest, usage of colour information from that region to yield a pulse signal and analysis of that signal to estimate the heart rate. This contribution proposes a novel approach to yield a region of interest using a Gaussian mixture model based level set formulation. The proposed method aims to segment a homogeneous region on an individual basis. To that end, we model the probability distributions for the pixel skin and non-skin class by two separate Gaussian mixture models. The proportion of the posterior probabilities are then included in the formulation of the level set function. The procedure yields a region of interest, which is used to derive a pulse signal from its average intensity or additional processing steps. We tested the method on own data and data of the 1st Challenge on remote Physiological Signal sensing. It is shown that the proposed method can improve the results for heart rate estimation on moving subjects. The potential of our approach is underlined by the promising result in the challenge.
In the present paper, classification and analysis of complex land features on the combined outcome of the high spectral resolution Hyperion-Hyperspectral remotesensing (HRS) data and high spatial resolution Resources...
详细信息
暂无评论