Classifying Forest Type in the National Forest Inventory Context with Airborne Hyperspectral and Lidar Data

Shoot, Caileigh; Andersen, Hans Erik; Moskal, L. Monika; Babcock, Chad; Cook, Bruce D.; Morton, Douglas C.

doi:10.3390/rs13101863

Cited by 21 publications

(10 citation statements)

References 61 publications

(56 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Topographic feature variables. The elevation of the study area (the data used are AW3D30 DSM) and its extracted slope, aspect, and slope position [ 86 ] were added to the feature variables.…”

Section: Methodsmentioning

confidence: 99%

Identification of Bamboo Species Based on Extreme Gradient Boosting (XGBoost) Using Zhuhai-1 Orbita Hyperspectral Remote Sensing Imagery

Zhou

Zhao

et al. 2022

Sensors

View full text Add to dashboard Cite

Mapping the distribution of bamboo species is vital for the sustainable management of bamboo and for assessing its ecological and socioeconomic value. However, the spectral similarity between bamboo species makes this work extremely challenging through remote sensing technology. Existing related studies rarely integrate multiple feature variables and consider how to quantify the main factors affecting classification. Therefore, feature variables, such as spectra, topography, texture, and vegetation indices, were used to construct the XGBoost model to identify bamboo species using the Zhuhai-1 Orbita hyperspectral (OHS) imagery in the Southern Sichuan Bamboo Sea and its surrounding areas in Sichuan Province, China. The random forest and Spearman’s rank correlation analysis were used to sort the main variables that affect classification accuracy and minimize the effects of multicollinearity among variables. The main findings were: (1) The XGBoost model achieved accurate and reliable classification results. The XGBoost model had a higher overall accuracy (80.6%), kappa coefficient (0.708), and mean F1-score (0.805) than the spectral angle mapper (SAM) method; (2) The optimal feature variables that were important and uncorrelated for classification accuracy included the blue band (B1, 464–468 nm), near-infrared band (B27, 861–871 nm), green band (B5, 534–539 nm), elevation, texture feature mean, green band (B4, 517–523 nm), and red edge band (B17, 711–720 nm); and (3) the XGBoost model based on the optimal feature variable selection showed good adaptability to land classification and had better classification performance. Moreover, the mean F1-score indicated that the model could well balance the user’s and producer’s accuracy. Additionally, our study demonstrated that OHS imagery has great potential for land cover classification and that combining multiple features to enhance classification is an approach worth exploring. Our study provides a methodological reference for the application of OHS images for plant species identification.

show abstract

Section: Methodsmentioning

confidence: 99%

Identification of Bamboo Species Based on Extreme Gradient Boosting (XGBoost) Using Zhuhai-1 Orbita Hyperspectral Remote Sensing Imagery

Zhou

Zhao

et al. 2022

Sensors

View full text Add to dashboard Cite

show abstract

“…Hyperspectral sensors are sensitive to even small spectral variations. This makes it possible to detect subtle changes in forestry (Jiang et al, 2021;Wang et al, 2019;Qin et al, 2022;Vangi et al, 2021;Shoot et al, 2021), agriculture (Lu et al, 2020;Hsieh et al,2020;Khan et al, 2022;9. Agilandeeswari et al, 2022;Pascucci et al, 2020), land cover (Ari, 2023;Yadav et al, 2022;Bhosle and Musande, 2019;Su et al, 2021, Ma et al, 2020Yang et al, 2022), material surfaces and other tasks (Kuras et al, 2021;Peyghambari and Zhang, 2021;Atik, 2022;Wang et al, 2021) that may not be distinguishable in standard RGB images.…”

Section: Introductionmentioning

confidence: 99%

Evaluation of CNN Models Using Deep Reinforcement Learning for Band Selection on Hyperspectral Image Classification

Atik

2023

Preprint

View full text Add to dashboard Cite

Along with the high spectral rich information it provides, one of the difficulties in processing a hyperspectral image is the need for expert knowledge and high-spec hardware to process very high-dimensional data. The use of the most relevant bands in the hyperspectral image is quite decisive in deep CNN networks without loss of information and loss of accuracy. It is crucial to classify hyperspectral images with faster and less hardware-requiring models by creating subset groups by choosing a limited number of optimal bands. In this study, a comparative analysis about the effect of deep reinforcement learning (DRL)-based hyperspectral band selection on the classification performance of deep learning networks is presented. 3D CNN, 3D + 1D CNN and Multiscale 3D deep convolutional neural network (M3D-DCNN) algorithms were used for hyperspectral image classification. By choosing the most effective bands determined by DRL, it is aimed to perform classification with high accuracy with fewer bands instead of all bands. All tests were performed on popular hyperspectral datasets, Indian Pines, Salinas, and Pavia Center. The 3D + 1D approach reached 92.28% OA in the IP dataset. In Salinas, 94.87% OA with 3D CNN and 94.62% OA with M3D-DCNN was obtained. 3D + 1D CNN has 98.64% OA in PaviaC.

show abstract

“…For many years, scientists have been conducting research related to the application of airborne [1][2][3] and satellite [4][5][6][7][8][9] remote sensing. Due to such applications, it is possible to study the tree stands and condition of vegetation in large areas without human intervention in the natural environment [10][11][12][13]. Since 2015, multispectral images of Sentinel-2 (S-2), Copernicus program are available with very good temporal (up to 2-3 days for middle latitudes) and spatial (10 m, 20 m) resolutions [14].…”

Section: Introductionmentioning

confidence: 99%

Forest Community Mapping Using Hyperspectral (CHRIS/PROBA) and Sentinel-2 Multispectral Images

Głowienka

Zembol²

2022

GaEE

View full text Add to dashboard Cite

The possibility to use hyperspectral images (CHRIS/PROBA) and multispectral images (Sentinel-2) in the classification of forest communities is assessed in this article. The pre-processing of CHRIS/PROBA image included: noise reduction, radiometric correction, atmospheric correction, geometric correction. Due to MNF transformation the number of the hyperspectral image channels was reduced (to 10 channels) and smiling errors were removed. Sentinel-2 image (level 2A) did not require pre-processing. Three tree genera occurring in the study area were selected for the classification: pine (Pinus), alder (Alnus) and birch (Betula). Image classification was carried out with three methods: SAM (Spectral Angle Mapper ), MTMF (Mixture Tuned Matched Filtering), SVM (Support Vector Machine). For the CHRIS/PROBA image, the algorithm SVM turned out to be the best. Its overall accuracy (OA) was 72%. The poorest result (OA = 52%) was for the MTMF classifier. In the classification of Sentinel-2 multispectral image the best result was for the MTMF method: OA = 82%, kappa coefficient 0.7. For other methods, the overall accuracy exceeded 65%. Among the classified genera, the highest producer’s accuracy was obtained for pine (PA = 96%), and the broad-leaf genera: alder and birch had PA ranging from 42% to 85%.

show abstract

Classifying Forest Type in the National Forest Inventory Context with Airborne Hyperspectral and Lidar Data

Cited by 21 publications

References 61 publications

Identification of Bamboo Species Based on Extreme Gradient Boosting (XGBoost) Using Zhuhai-1 Orbita Hyperspectral Remote Sensing Imagery

Identification of Bamboo Species Based on Extreme Gradient Boosting (XGBoost) Using Zhuhai-1 Orbita Hyperspectral Remote Sensing Imagery

Evaluation of CNN Models Using Deep Reinforcement Learning for Band Selection on Hyperspectral Image Classification

Forest Community Mapping Using Hyperspectral (CHRIS/PROBA) and Sentinel-2 Multispectral Images

Contact Info

Product

Resources

About