Search

Plant pests are the primary biological threats to agricultural and forestry production as well as forest ecosystem. Monitoring forest-pest damage via satellite images is crucial for the development of prevention and control strategies. Previous studies utilizing deep learning to monitor pest-infested damage in satellite imagery adopted RGB images, while multispectral imagery and vegetation indices were not used. Multispectral images and vegetation indices contain a wealth of useful information for detecting plant health, which can improve the precision of pest damage detection. The aim of the study is to further improve forest-pest infestation area segmentation by combining multispectral, vegetation indices and RGB information into deep learning. We also propose a new image segmentation method based on UNet++ with attention mechanism module for detecting forest damage induced by bark beetle and aspen leaf miner in Sentinel-2 images. The ResNeSt101 is used as the feature extraction backbone, and the attention mechanism scSE module is introduced in the decoding phase for improving the image segmentation results. We used Sentinel-2 imagery to produce a dataset based on forest health damage data gathered by the Ministry of Forests, Lands, Natural Resource Operations and Rural Development (FLNRORD) in British Columbia (BC), Canada, during aerial overview surveys (AOS) in 2020. The dataset contains the 11 original Sentinel-2 bands and 13 vegetation indices. The experimental results confirmed that the significance of vegetation indices and multispectral data in enhancing the segmentation effect. The results demonstrated that the proposed method exhibits better segmentation quality and more accurate quantitative indices with overall accuracy of 85.11%, in comparison with the state-of-the-art pest area segmentation methods. [ABSTRACT FROM AUTHOR]

In this paper, the main objective is to detect changes in the geographical area of Ottawa city in Canada due to floods. Two multi-temporal Synthetic Aperture Radar (SAR) images have been taken to evaluate the un-supervised change detection process. In this process, two ratio operators named as Log-Ratio and Mean-Ratio are used to generate a difference image. Performing image fusion based on DWT by selecting optimum filter coefficients by satisfying the wavelet filter coefficient properties through a novel image fusion technique is named as ADWT. GA, PSO, AntLion Optimization algorithms (ALO) and Hybridized AntLion Algorithm (HALO) have been adapted to perform the ADWT based image fusion. Segmentation has been performed based on fuzzy c-Means clustering to detect changed and unchanged pixels. Finally, the performance of the proposed method will be analysed by comparing the segmented image with the ground truth image in terms of sensitivity, accuracy, specificity, precision, F1-score. [ABSTRACT FROM AUTHOR]

Areas affected by forest fires are increasing worldwide, making salvage logging (i.e., harvesting fire-affected trees) an increasingly used practice to reduce the economic impacts of fire on forestry. However, salvage logging can have strong ecological impacts, notably on post-fire forest regeneration and biodiversity. Burn severity (i.e., the degree to which fires impact the vegetation and soil) is also a central element that interacts with pre-fire forest characteristics and salvages logging to control post-fire forest dynamics and biodiversity. In an ecosystem-based forest management context, spatial patterns of burn severity should thus be considered when planning salvage logging operations. This study presents a simple and innovative method to generate burn severity maps with Landsat and Sentinel-2 multispectral imageries to support salvage logging operations rapidly after a fire event. We assembled a unique dataset involving 330 plots from 10 burns (from 2010 to 2020) in which burn severity has been estimated on the field using the composite burn index (CBI) approach in eastern North America. CBI values were modelled as a univariate function of changes in spectral indices using the first cloud-free post-fire satellite images taken after the burns. Our results demonstrate that using free satellite images with straightforward methods can produce reliable and ecologically meaningful burn severity maps within the few weeks following a fire event. The method was then applied to a case study with salvage logging that illustrates how our burn severity maps could be a useful tool for guiding post-fire forestry operations in an ecosystem-based management context. We combined burn severity maps with pre-fire forest composition and age maps to assess immediate post-fire forest. We discuss how such an approach helps to guide salvage planning and maintain residual forests that are representative of the initial post-fire spatial variability in burn severity and pre-fire vegetation. We conclude that rapid mapping of burn severity after a forest fire event may offer many other applications for identifying and managing recently burned forests. • Interactions between salvage logging and burn severity control post-fire dynamics. • Unsalvaged residual forests should represent the variability in burn severity. • Free satellite imagery can rapidly produce burn severity maps after a fire event. • Such maps are powerful tool for guiding post-fire salvage logging operations. [ABSTRACT FROM AUTHOR]

Due to the advent of powerful parallel processing tools, including modern Graphics Processing Units (GPU), new deep learning algorithms, such as Convolutional Neural Networks (CNNs), have significantly altered the state-of-the-art algorithms in satellite classification of complex environments. Recent studies have demonstrated that the generic feature maps extracted from CNNs are incredibly effective in wetland classification. The main drawback of very deep CNNs is described as structurally complex, causing the need for extensive training data. To address deep Convolutional Neural Network's limitations, a timely and computationally efficient CNN architecture is proposed in this paper. The results of the proposed model were compared to other well-known CNNs (i.e., GoogleNet and SqueezeNet) and several machine learning algorithms, including Random Forest, Gaussian Naïve Bayes, and the Bayesian Optimized Tree. Results showed while significantly reduced the training time, the proposed deep learning method outperformed GoogleNet and SqueezeNet by about 12.71% and 12.2% in terms of mean overall accuracy, respectively. The classification results shown that the accuracy of wetland classes (fen, marsh, swamp, and shallow water) were significantly improved by applying the proposed CNN method. [ABSTRACT FROM AUTHOR]

In this article, we present an approach to land-use and land-cover (LULC) mapping from multispectral satellite images using deep learning methods. The terms satellite image classification and map production, although used interchangeably have specific meanings in the field of remote sensing. Satellite image classification describes assignment of global labels to entire scenes, whereas LULC map production involves producing maps by assigning a class to each pixel. We show that by classifying each pixel in a satellite image into a number of LULC categories we are able to successfully produce LULC maps. This process of LULC mapping is achieved using deep neural networks pre-trained on the ImageNet large-scale visual recognition competition datasets and fine-tuned on our target dataset, which consists of Landsat 5/7 multispectral satellite images taken of the Province of Manitoba in Canada. This approach resulted in 88% global accuracy. Performance was further improved by considering the state-of-the-art generative adversarial architecture and context module integrated with the original networks. The result is an automated deep learning framework that can produce highly accurate LULC maps images significantly faster than current semi-automated methods. The contribution of this article includes extensive experimentation of different FCN architectures with extensions on a unique dataset, high classification accuracy of 90.46%, and a thorough analysis and accuracy assessment of our results. [ABSTRACT FROM AUTHOR]

The fate of the carbon stored in frozen peatlands in Canada's Hudson Bay Lowlands (HBL) depends strongly on water, with wetter conditions favoring long-term storage. Dynamic surface water products generated from historical Landsat data exist to inform surface water trends in the HBL based on binary classifications of land vs water at 30 m spatial resolution. However, the HBL contains many water features smaller than 30 m, including streams and patterned fens that require a sub-pixel mapping approach to fully capture their dynamics. In this paper, we leverage an existing binary dynamic surface water product to generate a spatially and temporally comprehensive Landsat sub-pixel surface water time-series over the HBL. We first generate Landsat Red, NIR and SWIR composites from 1985 to 2021 and evaluate data-driven machine learning and physical models to derive sub-30 m water fractions across the entire mapping domain. Data-driven models were calibrated on 30 m water fractions obtained from 16 of 17 WorldView scene water fraction maps and assessed on the 17th held-out scene in a jackknife validation. Data-driven models were also calibrated by relating scaled Landsat reflectance to water fractions obtained from binary water masks at corresponding scales from 90 to 270 m. Physical models involved simple linear unmixing using a new approach that samples land endmembers in the vicinity of each pixel to represent land reflectance, and a global water endmember sampled beneath permanent water in each image. The assumption of linear scaling was verified at sub-30 m to 270 m resolution for NIR, SWIR and NDVI, which were all highly correlated with water fraction and had no significant differences among water fraction – spectral feature relationships across the full range of scales. Each of 11 different methods and input spectral features were evaluated against WorldView water fractions and ranked based on nine criteria. We found that machine learning using all spectral features including local NIR land endmembers performed best among data-driven approaches, and linear unmixing of the NIR band using local land endmembers performed best among physical. Both best-performing methods were applied to the entire time-series and surface water area trends were calculated and compared with trends obtained from binary water masks. All three methods agree on the direction of surface water change towards wetting, with sub-pixel inter-annual variation agreeing with stream discharge noted in other studies. • Physical and data-driven Landsat water fraction methods tested in the Hudson Bay Lowlands. • Physical methods perform linear unmixing between local land and global water endmembers. • Data-driven methods consist of machine learning calibrated using high-resolution water. • Best methods applied to 1985–2021 Landsat time-series reveal overall wetting trends. • Spatial trends generally suggest drying in bogs and wetting in fens. [ABSTRACT FROM AUTHOR]

Although research with digital airborne remote sensing data has been undertaken in different ecoregions to classify forested areas, the potential role of such imagery in deriving information to assist forest management has not yet been fully defined. The objective of this study was to determine the extent that the addition of texture could improve spectral classification of high spatial resolution images (pixel size 1m). These images represented pure and mixed wood forest stands from ecoregions in Alberta and New Brunswick, Canada. This study employed a judicious, selective application of texture to stands within a hierarchical classification framework. In Alberta, the addition of texture made a modest improvement in classification accuracy from 60% to 65%. In New Brunswick, the application of texture to selected land cover types resulted in an overall 12% improvement in classification accuracy. The addition of image texture increased classification accuracy for high spatial detail imagery relative to low spatial detail imagery. Incorporating texture into classification also improved classification accuracies for hardwood stands more so than for softwood stands, but greater attention to stand structure and composition will be needed in future work. Classification accuracies on the order of 60-65% were achieved with simple texture derivatives, maximum likelihood decision rules and conventional classification methods. [ABSTRACT FROM AUTHOR]

Multispectral/hyperspectral imaging is one of the imaging modalities to visualize and quantify blood supply in surface tissues such as skin or mucosa. The results of visualization can be potentially affected by various factors, for instance by elevated methemoglobin (MetHb) content (e.g., methemoglobinemia). The scope of the current study is to develop a robust approach for fabrication and validation of tissue-mimicking phantoms, which can be used to assess and improve tissue oximetry., Methods: The realistic tissue mimicking gelatin-based phantoms with intralipid (4% v/v) and/or hemoglobins (oxy-, deoxyhemoglobins, and MetHb) were molded between two coverslips separated by 2-mm wires. The hemoglobin solutions were prepared by dissolving the lyophilized human hemoglobin powder (H7379, Sigma-Aldrich) in the deionized water. Sodium dithionite (85% purity, 157,953, Sigma-Aldrich) was used to reduce MetHb solution. The phantoms were imaged using a multispectral imaging device (Oxilight, Canada).To demonstrate the utility, the developed approach is applied to emulate elevated systemic MetHb content., Results: Initial results show that elevated systemic MetHb (2.0-6.7% of total blood) does not impact the accuracy of tissue oximetry imaging., Discussion: A robust method for fabrication and optical validation of biocompatible tissue-mimicking phantoms has been developed.The proposed phantom design allows combining different phantoms into multilayer (sandwich) structures, which can be used to emulate a wide range of topical and systemic conditions.

Time series of vegetation biophysical variables (leaf area index (LAI), fraction canopy cover (FCOVER), fraction of absorbed photosynthetically active radiation (FAPAR), canopy chlorophyll content (CCC), and canopy water content (CWC)) were estimated from interpolated Sentinel-2 (S2-LIKE) surface reflectance images, for an agricultural region located in central Canada, using the Simplified Level 2 Product Prototype Processor (SL2P). S2-LIKE surface reflectance data were generated by blending clear-sky Sentinel-2 Multispectral Imager (S2-MSI) images with daily BRDF-adjusted Moderate Resolution Imaging Spectrometer images using the Prediction Smooth Reflectance Fusion Model (PSFRM), and validated using thirteen independent S2-MSI images (RMSE ≤ 6%). The uncertainty of S2-LIKE surface reflectance data increases with the time delay between the prediction date and the closest S2-MSI image used for training PSFRM. Vegetation biophysical variables from S2-LIKE products are validated qualitatively and quantitatively by comparison to the corresponding vegetation biophysical variables from S2-MSI products (RMSE = 0.55 for LAI, ~10% for FCOVER and FAPAR, and 0.13 g/m2 for CCC and 0.16 kg/m2 for CWC). Uncertainties of vegetation biophysical variables derived from S2-LIKE products are almost linearly related to the uncertainty of the input reflectance data. When compared to the in situ measurements collected during the Soil Moisture Active Passive Validation Experiment 2016 field campaign, uncertainties of LAI (0.83) and FCOVER (13.73%) estimates from S2-LIKE products were slightly larger than uncertainties of LAI (0.57) and FCOVER (11.80%) estimates from S2-MSI products. However, equal uncertainties (0.32 kg/m2) were obtained for CWC estimates using SL2P with either S2-LIKE or S2-MSI input data. [ABSTRACT FROM AUTHOR]

Despite recent advances of deep Convolutional Neural Networks (CNNs) in various computer vision tasks, their potential for classification of multispectral remote sensing images has not been thoroughly explored. In particular, the applications of deep CNNs using optical remote sensing data have focused on the classification of very high-resolution aerial and satellite data, owing to the similarity of these data to the large datasets in computer vision. Accordingly, this study presents a detailed investigation of state-of-the-art deep learning tools for classification of complex wetland classes using multispectral RapidEye optical imagery. Specifically, we examine the capacity of seven well-known deep convnets, namely DenseNet121, InceptionV3, VGG16, VGG19, Xception, ResNet50, and InceptionResNetV2, for wetland mapping in Canada. In addition, the classification results obtained from deep CNNs are compared with those based on conventional machine learning tools, including Random Forest and Support Vector Machine, to further evaluate the efficiency of the former to classify wetlands. The results illustrate that the full-training of convnets using five spectral bands outperforms the other strategies for all convnets. InceptionResNetV2, ResNet50, and Xception are distinguished as the top three convnets, providing state-of-the-art classification accuracies of 96.17%, 94.81%, and 93.57%, respectively. The classification accuracies obtained using Support Vector Machine (SVM) and Random Forest (RF) are 74.89% and 76.08%, respectively, considerably inferior relative to CNNs. Importantly, InceptionResNetV2 is consistently found to be superior compared to all other convnets, suggesting the integration of Inception and ResNet modules is an efficient architecture for classifying complex remote sensing scenes such as wetlands. [ABSTRACT FROM AUTHOR]