Radiometric normalization is a fundamental preprocessing step for the analysis of multi-temporal and multi-sensor satellite images. Some of previous radiometric normalization studies addressed the elimination of spectral differences between bitemporal or cross-sensor images by performing canonical correlation analysis (CCA) or its generalization called kernel CCA (KCCA). The KCCA-based normalization methods, which assume that the relation between at-sensor radiances is spatially nonlinear, can successfully extract pseudo-invariant features (PIFs) for bitemporal images with nonlinear differences caused by seasonal variation and for bitemporal images acquired by different sensors with different spectral centers and bandwidths of spectral bands. However, these methods did not consider the problems of high computation and storage costs induced by the huge size of kernel matrices and overfitting problem caused by non-linear regression, making the methods infeasible for satellite images of large sizes. To solve these problems, a hybrid CCA with hybrid regression model is proposed, which can preserve the characteristics of nonlinear CCA and regression while reducing time and space complexity and solving overfitting problems. Cross-sensor images acquired by Landsat-7 Enhanced Thematic Mapper Plus and Landsat-8 Operational and Imager sensors were used to evaluate the proposed method. Experimental results demonstrate the superiority of the proposed method to the related CCA-based methods in terms of computational performance and PIF quality.

Deteksi dini kerusakan jalan, seperti retakan dan deformasi lainnya, sangat penting karena berfungsi sebagai indikatorawal keausan, kerusakan, atau masalah struktural yang mendasari yang dapat membahayakan keselamatan dan daya tahaninfrastruktur jalan. Penelitian ini bertujuan untuk mengembangkan sistem inspeksi otomatis berbasis pembelajaran mendalam(deep learning) dan pemrosesan citra untuk mendeteksi dan mengklasifikasikan kerusakan jalan secara efektif. Dalam proyek ini,kami menggunakan model pembelajaran mendalam dengan CNN (Convolutional Neural Networks) dan arsitektur ResNet untukmempelajari fitur intrinsik kerusakan permukaan jalan melalui gambar, yang kemudian digunakan untuk klasifikasi otomatis kedalam kelas rusak/tidak rusak atau deformasi permukaan. Model YOLO ini menggunakan metode pembelajaran transfer daridata global dan untuk data jalan nasional yang ada di Indonesia. Tidak seperti pendekatan konvensional, penelitian inimenerapkan model YOLO (You Only Look Once) untuk mendeteksi dan mengklasifikasikan kerusakan jalan secara real-time,memungkinkan deteksi langsung pada berbagai permukaan jalan. Dengan kemampuan YOLO dalam deteksi objek yang cepatdan efisien, sistem ini dapat digunakan dalam inspeksi infrastruktur jalan secara rutin dan pemantauan kondisi jalan secaraberkelanjutan. Pendekatan ini telah dievaluasi menggunakan dataset kerusakan jalan yang tersedia di jalan nasional di Palembang,Sumatera Selatan, Indonesia, dan hasil evaluasi menunjukkan bahwa model YOLO tidak hanya mendeteksi kerusakan jalandengan akurasi tinggi, tetapi juga menunjukkan kecepatan deteksi yang unggul dibandingkan dengan metode tradisional. Hasilpenelitian ini menunjukkan potensi signifikan penggunaan YOLO untuk inspeksi jalan otomatis, membantu mengidentifikasi danmenangani kerusakan dini serta mendukung upaya pemeliharaan dan keselamatan jalan.

Surface reflectance derived from optical satellite images is a fundamental and important material for remote sensing applications. The advanced atmospheric corrections have achieved a great progress in removal of atmospheric effects by using complicated image- and physical-based models. Those algorithms are processed in each image data acquisition and using additional in-situ data information. However, the process of image-by-image atmospheric correction may cause biases for multitemporal images, which produce temporal inconsistencies and discontinuities over the surface reflectance of invariant ground objects. In this study, a surface reflectance bias removal is proposed for multispectral and multitemporal satellite images using a constrained harmonization. The constrained harmonization is applied by using pseudoinvariant features (PIFs) instead of an entire image. The PIFs are determined by using iterative reweighted multitemporal and multivariate alteration detection (IRMMAD). With using PIFs, the removal of surface reflectance bias in multispectral and multitemporal images is achieved through solving a constrained least-squares linear system simultaneously. Those constraints are called temporal and spectral consistency constraints which can maintain the temporal and spectral consistencies, respectively. To test the robustness of the proposed method, surface reflectance derived from six atmospheric correction (AC) models were tested. Qualitative and quantitative analyses on the surface reflectance of multitemporal images have been conducted. The experimental results show that the surface reflectance biases can be significantly removed while the spectral signatures can be preserved. In addition, the proposed method outperforms related paired-wise harmonization and temporal-constrained harmonization methods in terms of consistency measurements.

The use of hyperspectral reflectance and transmittance data, as well as advanced machine learning approaches, has substantially improved material analysis by expanding into the near-infrared (NIR) and shortwave infrared (SWIR) areas beyond the visible spectrum. This study looks into fresh green tea leaves’ SWIR-2 region (2001–2500 nm), which is important for materials science, agriculture, and environmental monitoring applications. While hyperspectral data contains abundant and detailed information, its complexity and high dimensionality pose difficulties for precise analysis. We use deep learning methods, particularly convolutional neural networks (CNNs)-based dual-input/output (Dual-Net), to address these issues by refining spectral signatures and improving prediction accuracy across a variety of applications. Our method accounts for spectral dependencies, improving the accuracy of predictions concerning leaf attributes like water content, nitrogen and chlorophyll concentrations, and other biophysical properties, even under complex environmental conditions. The findings show that deep learning models make effective use of SWIR-2 spectra for important evaluations of plant health. The proposed method produced mean (±SD) root-mean-square errors (RMSEs) of 0.012 (±0.012) of reflectance and 0.011 (±0.010) of transmittance for individual spectra, and RMSEs of 0.028 (±0.029) of reflectance and 0.022 (±0.023) of transmittance for SWIR-2 bands. This method shows strong potential for advancing remote sensing applications, particularly in monitoring plant health and productivity. Further improvements could be realized through the development of a more comprehensive green leaf spectral database.

Twitter has become one of the most significant resources for text mining. Twitter can provide information about human activities, mobility, and emotional patterns along with location data. Many types of text research can be made with these data, one of which is sentiment analysis. This study evaluates the potential of deriving emotional responses of individuals from tweets while they experience and interact with urban green space. A machine learning model using Support Vector Machine (SVM) and corpus from over 2000 movie reviews has been made. This model is used to classify incoming tweets into positive and negative sentiments. Then the web-based recommender system has been built to provide suggestions for green spaces based on users' preferred activities.