Leveraging Machine Learning to Extend Ontology-Driven Geographic Object-Based Image Analysis (O-GEOBIA): A Case Study in Forest-Type Mapping

Active Actions in the Extraction of Urban Objects for Information Quality and Knowledge Recommendation with Machine Learning

Due to the increasing urban development, it has become important for municipalities to permanently understand land use and ecological processes, and make cities smart and sustainable by implementing technological tools for land monitoring. An important problem is the absence of technologies that certify the quality of information for the creation of strategies. In this context, expressive volumes of data are used, requiring great effort to understand their structures, and then access information with the desired quality. This study are designed to provide an initial response to the need for mapping zones in the city of Itajaí (SC), Brazil. The solution proposes to aid object recognition employing object-based classifiers OneR, NaiveBayes, J48, IBk, and Hoeffding Tree algorithms used together with GeoDMA, and a first approach in the use of Region-based Convolutional Neural Network (R-CNN) and the YOLO algorithm. All this is to characterize vegetation zones, exposed soil zones, asphalt, and buildings within an urban and rural area. Through the implemented model for active identification of geospatial objects with similarity levels, it was possible to apply the data crossover after detecting the best classifier with accuracy (85%) and the kappa agreement coefficient (76%). The case study presents the dynamics of urban and rural expansion, where expressive volumes of data are obtained and submitted to different methods of cataloging and preparation to subsidize rapid control actions. Finally, the research describes a practical and systematic approach, evaluating the extraction of information to the recommendation of knowledge with greater scientific relevance. Allowing the methods presented to apply the calibration of values for each object, to achieve results with greater accuracy, which is proposed to help improve conservation and management decisions related to the zones within the city, leaving as a legacy the construction of a minimum technological infrastructure to support the decision.

Analysis and comparison of machine learning methods for blood identification using single-cell laser tweezer Raman spectroscopy

Raman spectroscopy, a "fingerprint" spectrum of substances, can be used to characterize various biological and chemical samples. To allow for blood classification using single-cell Raman spectroscopy, several machine learning algorithms were implemented and compared. A single-cell laser optical tweezer Raman spectroscopy system was established to obtain the Raman spectra of red blood cells. The Boruta algorithm extracted the spectral feature frequency shift, reduced the spectral dimension, and determined the essential features that affect classification. Next, seven machine learning classification models are analyzed and compared based on the classification accuracy, precision, and recall indicators. The results show that support vector machines and artificial neural networks are the two most appropriate machine learning algorithms for single-cell Raman spectrum blood classification, and this finding provides essential guidance for future research studies.

Automated Delineation of Microstands in Hemiboreal Mixed Forests Using Stereo GeoEye-1 Data

A microstand is a small forest area with a homogeneous tree species, height, and density composition. High-spatial-resolution GeoEye-1 multispectral (MS) images and GeoEye-1-based canopy height models (CHMs) allow delineating microstands automatically. This paper studied the potential benefits of two microstand segmentation workflows: (1) our modification of JSEG and (2) generic region merging (GRM) of the Orfeo Toolbox, both intended for the microstand border refinement and automated stand volume estimation in hemiboreal forests. Our modification of JSEG uses a CHM as the primary data source for segmentation by refining the results using MS data. Meanwhile, the CHM and multispectral data fusion were achieved as multiband segmentation for the GRM workflow. The accuracy was evaluated using several sets of metrics (unsupervised, supervised direct assessment, and system-level assessment). Metrics were calculated for a regular segment grid to check the benefits compared with the simple image patches. The metrics showed very similar results for both workflows. The most successful combinations in the workflow parameters retrieved over 75 % of the boundaries selected by a human interpreter. However, the impact of data fusion and parameter combinations on stand volume estimation accuracy was minimal, causing variations of the RMSE within approximately 7 m3/ha.

Potential of Vis-NIR to measure heavy metals in different varieties of organic-fertilizers using Boruta and deep belief network

The quick identification of heavy metals is of major importance and is beneficial for controlling the fertilizer production process in the fertilizer industries. This work aimed to use visible and near-infrared spectroscopy (Vis-NIR), Boruta, and deep learning to establish rapid heavy metals screening methods. Boruta algorithm was used to extract appropriate wavelengths, and a deep belief network (DBN) was computed to determine the amounts of various heavy metals such as chromium (Cr), cadmium (Cd), lead (Pb), and mercury (Hg) for both the entire and selected wavelengths. To assess the model, coefficient of determination (R2), root mean squared error (RMSE), and residual prediction deviation (RPD) were used to calculate the reliability of the model. The results of the selected wavelengths were excellent and much higher than the full wavelengths with R2p = 0.96, RMSEP = 0.2017 mg kg-1 and RPDpred = 5.0 for Cr; R2p = 0.91, RMSEP = 0.2832 mg kg-1 and RPDpred = 3.4 for Pb; R2p = 0.90, RMSEP = 0.2992 mg kg-1, and RPDpred = 3.3 for Hg. Descent prediction was obtained also for Cd (R2p = 0.87, RMSEP = 0.3435 mg kg-1, and RPDpred = 2.7). To further assess the robustness of the DBN, it was compared with conventional machine learning methods such as support vector machine for regression (SVR), k nearest neighbor (KNN), and partial least squares (PLS). The overall results indicated that the Vis-NIR technique coupled with Boruta and DBN could be reliable and accurate for screening heavy metals in organic fertilizers.

Object Oriented Classification for Mapping Mixed and Pure Forest Stands Using Very-High Resolution Imagery

The importance of mixed forests is increasingly recognized on a scientific level, due to their greater productivity and efficiency in resource use, compared to pure stands. However, a reliable quantification of the actual spatial extent of mixed stands on a fine spatial scale is still lacking. Indeed, classification and mapping of mixed populations, especially with semi-automatic procedures, has been a challenging issue up to date. The main objective of this study is to evaluate the potential of Object-Based Image Analysis (OBIA) and Very-High-Resolution imagery (VHR) to detect and map mixed forests of broadleaves and coniferous trees with a Minimum Mapping Unit (MMU) of 500 m2. This study evaluates segmentation-based classification paired with non-parametric method K- nearest-neighbors (K-NN), trained with a dataset independent from the validation one. The forest area mapped as mixed forest canopies in the study area amounts to 11%, with an overall accuracy being equal to 85% and K of 0.78. Better levels of user and producer accuracies (85–93%) are reached in conifer and broadleaved dominated stands. The study findings demonstrate that the very high resolution images (0.20 m of spatial resolutions) can be reliably used to detect the fine-grained pattern of rare mixed forests, thus supporting the monitoring and management of forest resources also on fine spatial scales.

Landslide Detection Using Multi-Scale Image Segmentation and Different Machine Learning Models in the Higher Himalayas

Landslides represent a severe hazard in many areas of the world. Accurate landslide maps are needed to document the occurrence and extent of landslides and to investigate their distribution, types, and the pattern of slope failures. Landslide maps are also crucial for determining landslide susceptibility and risk. Satellite data have been widely used for such investigations—next to data from airborne or unmanned aerial vehicle (UAV)-borne campaigns and Digital Elevation Models (DEMs). We have developed a methodology that incorporates object-based image analysis (OBIA) with three machine learning (ML) methods, namely, the multilayer perceptron neural network (MLP-NN) and random forest (RF), for landslide detection. We identified the optimal scale parameters (SP) and used them for multi-scale segmentation and further analysis. We evaluated the resulting objects using the object pureness index (OPI), object matching index (OMI), and object fitness index (OFI) measures. We then applied two different methods to optimize the landslide detection task: (a) an ensemble method of stacking that combines the different ML methods for improving the performance, and (b) Dempster–Shafer theory (DST), to combine the multi-scale segmentation and classification results. Through the combination of three ML methods and the multi-scale approach, the framework enhanced landslide detection when it was tested for detecting earthquake-triggered landslides in Rasuwa district, Nepal. PlanetScope optical satellite images and a DEM were used, along with the derived landslide conditioning factors. Different accuracy assessment measures were used to compare the results against a field-based landslide inventory. All ML methods yielded the highest overall accuracies ranging from 83.3% to 87.2% when using objects with the optimal SP compared to other SPs. However, applying DST to combine the multi-scale results of each ML method significantly increased the overall accuracies to almost 90%. Overall, the integration of OBIA with ML methods resulted in appropriate landslide detections, but using the optimal SP and ML method is crucial for success.

Uncertainty Assessment in Multitemporal Land Use/Cover Mapping with Classification System Semantic Heterogeneity

Land use/cover (LUC) data are commonly relied on to provide land surface information in a variety of applications. However, the exchange and joint use of LUC information from different datasets can be challenging due to semantic differences between common classification systems (CSs). In this paper, we propose an uncertainty assessment schema to capture the semantic translation uncertainty between heterogeneous LUC CSs and evaluate the data label uncertainty of multitemporal LUC mapping results caused by uncertainty propagation. The semantic translation uncertainty between CSs is investigated using a dynamic semantic reference system (DSRS) model and semantic similarity analysis. An object-based unsupervised change detection algorithm is adopted to determine the probability of changes in land patches, and novel uncertainty metrics are proposed to estimate the patch label uncertainty in LUC maps. The proposed uncertainty assessment schema was validated via experiments on four LUC datasets, and the results confirmed that semantic uncertainty had great impact on data reliability and that the uncertainty metrics could be used in the development of uncertainty controls in multitemporal LUC mapping by referring to uncertainty assessment results. We anticipate our findings will be used to improve the applicability and interoperability of LUC data products.