Search Results (5)

This research presents DuSAFNet, a lightweight deep neural network for fine-grained bird audio classification. DuSAFNet combines dual-path feature fusion, spectral–temporal attention, and a multi-band ArcMarginProduct classifier to enhance inter-class separability and capture both local and global spectro–temporal cues. Unlike single-feature approaches, DuSAFNet captures both local spectral textures and long-range temporal dependencies in Mel-spectrogram inputs and explicitly enhances inter-class separability across low, mid, and high frequency bands. On a curated dataset of 17,653 three-second recordings spanning 18 species, DuSAFNet achieves 96.88% accuracy and a 96.83% F1 score using only 6.77 M parameters and 2.275 GFLOPs. Cross-dataset evaluation on Birdsdata yields 93.74% accuracy, demonstrating robust generalization to new recording conditions. Its lightweight design and high performance make DuSAFNet well-suited for edge-device deployment and real-time alerts for rare or threatened species. This work lays the foundation for scalable, automated acoustic monitoring to inform biodiversity assessments and conservation planning. Full article

Open areas, along with their non-forest vegetation, are often threatened by secondary succession, which causes deterioration of biodiversity and the habitat’s conservation status. The knowledge about characteristics and dynamics of the secondary succession process is very important in the context of management and proper planning of active protection of the Natura 2000 habitats. This paper presents research on the evaluation of the possibility of using selected methods of textural analysis to determine the spatial extent of trees and shrubs based on archival aerial photographs, and consequently on the investigation of the secondary succession process. The research was carried out on imagery from six different dates, from 1971 to 2015. The images differed from each other in spectral resolution (panchromatic, in natural colors, color infrared), in original spatial resolution, as well as in radiometric quality. Two methods of textural analysis were chosen for the analysis: Gray level co-occurrence matrix (GLCM) and granulometric analysis, in a number of variants, depending on the selected parameters of these transformations. The choice of methods has been challenged by their reliability and ease of implementation in practice. The accuracy assessment was carried out using the results of visual photo interpretation of orthophotomaps from particular years as reference data. As a result of the conducted analyses, significant efficacy of the analyzed methods has been proved, with granulometric analysis as the method of generally better suitability and greater stability. The obtained results show the impact of individual image features on the classification efficiency. They also show greater stability and reliability of texture analysis based on granulometric/morphological operations. Full article

The paper presents a comparison of the efficacy of several texture analysis methods as tools for improving land use/cover classification in satellite imagery. The tested methods were: gray level co-occurrence matrix (GLCM) features, Laplace filters and granulometric analysis, based on mathematical morphology. The performed tests included an assessment of the classification accuracy performed based on spectro-textural datasets: spectral images with the addition of images generated using different texture analysis methods. The class nomenclature was based on spectral and textural differences and included the following classes: water, low vegetation, bare soil, urban, and two (coniferous and deciduous) forest classes. The classification accuracy was assessed using the overall accuracy and kappa index of agreement, based on the reference data generated using visual interpretation of the images. The analysis was performed using very high-resolution imagery (Pleiades, WorldView-2) and high-resolution imagery (Sentinel-2). The results show the efficacy of selected GLCM features and granulometric analysis as tools for providing textural data, which could be used in the process of land use/cover classification. It is also clear that texture analysis is generally a more important and effective component of classification for images of higher resolution. In addition, for classification using GLCM results, the Random Forest variable importance analysis was performed. Full article

This paper presents a novel approach for acoustic scene classification based on efficient acoustic feature extraction using spectro-temporal descriptors fusion. Grounded on the finding in neuroscience—“auditory system summarizes the temporal details of sounds using time-averaged statistics to understand acoustic scenes”, we devise an efficient computational framework for sound scene classification by using multipe time-frequency descriptors fusion with discriminant information enhancement. To characterize rich information of sound, i.e., local structures on the time-frequency plane, we adopt 2-dimensional local descriptors. A more critical issue raised in how to logically ‘summarize’ those local details into a compact feature vector for scene classification. Although ‘time-averaged statistics’ is suggested by the psychological investigation, directly computing time average of local acoustic features is not a logical way, since arithmetic mean is vulnerable to extreme values which are anticipated to be generated by interference sounds which are irrelevant to the scene category. To tackle this problem, we develop time-frame weighting approach to enhance sound textures as well as to suppress scene-irrelevant events. Subsequently, robust acoustic feature for scene classification can be efficiently characterized. The proposed method had been validated by using Rouen dataset which consists of 19 acoustic scene categories with 3029 real samples. Extensive results demonstrated the effectiveness of the proposed scheme. Full article

The overarching goal of this research was to develop and demonstrate an automated Cropland Classification Algorithm (ACCA) that will rapidly, routinely, and accurately classify agricultural cropland extent, areas, and characteristics (e.g., irrigated vs. rainfed) over large areas such as a country or a region through combination of multi-sensor remote sensing and secondary data. In this research, a rule-based ACCA was conceptualized, developed, and demonstrated for the country of Tajikistan using mega file data cubes (MFDCs) involving data from Landsat Global Land Survey (GLS), Landsat Enhanced Thematic Mapper Plus (ETM+) 30 m, Moderate Resolution Imaging Spectroradiometer (MODIS) 250 m time-series, a suite of secondary data (e.g., elevation, slope, precipitation, temperature), and in situ data. First, the process involved producing an accurate reference (or truth) cropland layer (TCL), consisting of cropland extent, areas, and irrigated vs. rainfed cropland areas, for the entire country of Tajikistan based on MFDC of year 2005 (MFDC2005). The methods involved in producing TCL included using ISOCLASS clustering, Tasseled Cap bi-spectral plots, spectro-temporal characteristics from MODIS 250 m monthly normalized difference vegetation index (NDVI) maximum value composites (MVC) time-series, and textural characteristics of higher resolution imagery. The TCL statistics accurately matched with the national statistics of Tajikistan for irrigated and rainfed croplands, where about 70% of croplands were irrigated and the rest rainfed. Second, a rule-based ACCA was developed to replicate the TCL accurately (~80% producer’s and user’s accuracies or within 20% quantity disagreement involving about 10 million Landsat 30 m sized cropland pixels of Tajikistan). Development of ACCA was an iterative process involving series of rules that are coded, refined, tweaked, and re-coded till ACCA derived croplands (ACLs) match accurately with TCLs. Third, the ACCA derived cropland layers of Tajikistan were produced for year 2005 (ACL2005), same year as the year used for developing ACCA, using MFDC2005. Fourth, TCL for year 2010 (TCL2010), an independent year, was produced using MFDC2010 using the same methods and approaches as the one used to produce TCL2005. Fifth, the ACCA was applied on MFDC2010 to derive ACL2010. The ACLs were then compared with TCLs (ACL2005 vs. TCL2005 and ACL2010 vs. TCL2010). The resulting accuracies and errors from error matrices involving about 152 million Landsat (30 m) pixels of the country of Tajikistan (of which about 10 million Landsat size, 30 m, cropland pixels) showed an overall accuracy of 99.6% (k_hat = 0.97) for ACL2005 vs. TCL2005. For the 3 classes (irrigated, rainfed, and others) mapped in ACL2005, the producer’s accuracy was >86.4% and users accuracy was >93.6%. For ACL2010 vs. TCL2010, the error matrix showed an overall accuracy on 96.2% (k_hat = 0.96). For the 3 classes (irrigated, rainfed, and others) mapped in ACL2010, the producer’s and user’s accuracies for the irrigated areas were ≥82.9%. Any intermixing was overwhelmingly between irrigated and rainfed croplands, indicating that croplands (irrigated plus rainfed areas) as well as irrigated areas were mapped with high levels of accuracies (~90% or higher) even for the independent year. The ACL2005 and ACL2010, each, were produced using ACCA algorithm in ~30 min using a Dell Precision desktop T7400 computer for the entire country of Tajikistan once the MFDCs for the years were ready. The ACCA algorithm for Tajikistan is made available through US Geological Survey’s ScienceBase: http://www.sciencebase.gov/catalog/folder/4f79f1b7e4b0009bd827f548 or at: https://powellcenter.usgs.gov/globalcroplandwater/content/models-algorithms. The research contributes to the efforts of global food security through research on global croplands and their water use (e.g., https://powellcenter.usgs.gov/globalcroplandwater/). The above results clearly demonstrated the ability of a rule-based ACCA to rapidly and accurately produce cropland data layer year after year (hindcast, nowcast, forecast) for the country it was developed using MFDCs that consist of combining multiple sensor data and secondary data. It needs to be noted that the ACCA is applicable to the area (e.g., country, region) for which it is developed. In this case, ACCA is applicable for the Country of Tajikistan to hindcast, nowcast, and forecast agricultural cropland extent, areas, and irrigated vs. rainfed. The same fundamental concept of ACCA applies to other areas of the World where ACCA codes need to be modified to suite the area/region of interest. ACCA can also be expanded to compute other crop characteristics such as crop types, cropping intensities, and phenologies. Full article