Segmentation of Remote Sensing Images Based on U-Net Multi-Task Learning

Ni, Ruiwen; Ye, Mu; Ji, Li; Zhang, Tong; Tianye, Luo; Feng, Ruilong; Gong, He; Hu, Tianli; Sun, Yu; Guo, Ying; Li, Shijun; Tyasi, Thobela Louis

doi:10.32604/cmc.2022.026881

Cited by 3 publications

(2 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a method of deep learning, U-net is a segmentation network based on feature extraction evolved from the fully convolutional network (FCN). Compared with other neural networks, such as convolutional neural network (CNN) and FCN, U-net can solve the problems of less fuzzy segmentation results and longer training time in jumping layer structure by transforming the input size of the image [75,76]. Additionally, U-net can not only reflect the global information of the image, but also reflect the details of the image, which makes it possible to greatly improve the accuracy of image features [77].…”

Section: ) Image Feature Extraction Based On U-netmentioning

confidence: 99%

Using POI Data and Baidu Migration Big Data to Modify Nighttime Light Data to Identify Urban and Rural Area

Chen

Deng

2022

IEEE Access

View full text Add to dashboard Cite

The spatial difference between urban and rural areas is the direct result of urban-rural relations. Accurate identification of urban-rural area is helpful to judge the urban-rural mechanism and promote the integration development of urban-rural area. Previous studies only used single nighttime light (NTL) data to identify urban and rural areas, which is likely to have an impact on the identification results due to the large brightness difference of lights. Therefore, based on NTL data and combine with data level fusion algorithm, this study separately fuses point of interest (POI) data that representing the quantity distribution of urban infrastructure and Baidu migration big (BM)data that representing the change relationship of regional population mobility to identify urban and rural areas by using deep learning method. The results show that the highest accuracy of urban-rural spatial identification with single NTL data is 84.32% and kappa is 0.6952, while the highest accuracy identified by data fusion is 95.02% and kappa is 0.8259. It can be seen that the differences caused by light brightness are effectively corrected after data fusion, which greatly improves the accuracy of urban and rural spatial identification. By comparing the results of NTL data modified by different big data, this study analyzes and identifies the accuracy of urban and rural area by using deep learning method, which not only enriches the study of data fusion in urban area, but also provides a basis for analyzing regional urban-rural relations and urban-rural development. Therefore, this study is believed to have important practical value for the coordinated development of urban and rural areas.

show abstract

Section: ) Image Feature Extraction Based On U-netmentioning

confidence: 99%

Using POI Data and Baidu Migration Big Data to Modify Nighttime Light Data to Identify Urban and Rural Area

Chen

Deng

2022

IEEE Access

View full text Add to dashboard Cite

show abstract

“…There are different applications of DL methods for RSI, i.e., SC [5], semantic segmentation [6,7], object detection [8,9], and change detection [10]. SC is the fundamental application, owing to the fact that its feature extraction is the basis of the others.…”

Section: Introductionmentioning

confidence: 99%

FST-EfficientNetV2: Exceptional Image Classification for Remote Sensing

Song¹

2023

Computer Systems Science and Engineering

View full text Add to dashboard Cite

Recently, the semantic classification (SC) algorithm for remote sensing images (RSI) has been greatly improved by deep learning (DL) techniques, e.g., deep convolutional neural networks (CNNs). However, too many methods employ complex procedures (e.g., multi-stages), excessive hardware budgets (e.g., multi-models), and an extreme reliance on domain knowledge (e.g., handcrafted features) for the pure purpose of improving accuracy. It obviously goes against the superiority of DL, i.e., simplicity and automation. Meanwhile, these algorithms come with unnecessarily expensive overhead on parameters and hardware costs. As a solution, the author proposed a fast and simple training algorithm based on the smallest architecture of Effi-cientNet version 2, which is called FST-EfficientNet. The approach employs a routine transfer learning strategy and has fast training characteristics. It outperforms all the former methods by a 0.8%-2.7% increase in accuracy. It does, however, use a higher testing resolution of 5122 and 6002, which results in high consumption of graphics processing units (GPUs). As an upgrade option, the author proposes a novel and more efficient method named FST-EfficientNetV2 as the successor. The new algorithm still employs a routine transfer learning strategy and maintains fast training characteristics. But a set of crucial algorithmic tweaks and hyperparameter re-optimizations have been updated. As a result, it achieves a noticeable increase in accuracy of 0.3%-1.1% over its predecessor. More importantly, the algorithm's GPU costs are reduced by 75%-81%, with a significant reduction in training time costs of 60%-80%. The results demonstrate that an efficient training optimization strategy can significantly boost the CNN algorithm's performance for RSI-SC. More crucially, the results prove that the distribution shift introduced by data augmentation (DA) techniques is vital to the method's performance for RSI-SC, which has been ignored to date. These findings may help us gain a correct understanding of the CNN algorithm for RSI-SC.

show abstract

DenseNet weed recognition model combining local variance preprocessing and attention mechanism

et al. 2023

Front. Plant Sci.

View full text Add to dashboard Cite

IntroductionThe purpose of this paper is to effectively and accurately identify weed species in crop fields in complex environments. There are many kinds of weeds in the detection area, which are densely distributed.MethodsThe paper proposes the use of local variance pre-processing method for background segmentation and data enhancement, which effectively removes the complex background and redundant information from the data, and prevents the experiment from overfitting, which can improve the accuracy rate significantly. Then, based on the optimization improvement of DenseNet network, Efficient Channel Attention (ECA) mechanism is introduced after the convolutional layer to increase the weight of important features, strengthen the weed features and suppress the background features.ResultsUsing the processed images to train the model, the accuracy rate reaches 97.98%, which is a great improvement, and the comprehensive performance is higher than that of DenseNet, VGGNet-16, VGGNet-19, ResNet-50, DANet, DNANet, and U-Net models.DiscussionThe experimental data show that the model and method we designed are well suited to solve the problem of accurate identification of crop and weed species in complex environments, laying a solid technical foundation for the development of intelligent weeding robots.

show abstract

Segmentation of Remote Sensing Images Based on U-Net Multi-Task Learning

Cited by 3 publications

References 29 publications

Using POI Data and Baidu Migration Big Data to Modify Nighttime Light Data to Identify Urban and Rural Area

Using POI Data and Baidu Migration Big Data to Modify Nighttime Light Data to Identify Urban and Rural Area

FST-EfficientNetV2: Exceptional Image Classification for Remote Sensing

DenseNet weed recognition model combining local variance preprocessing and attention mechanism

Contact Info

Product

Resources

About