Building detection from different high-resolution aerial and satellite images has been a notable research topic in recent decades. The primary challenges are occlusions, shadows, different roof types, and similar spectral behavior of urban covers. Integration of different data sources is a solution to supplement the input feature space and improve the existing algorithms. Regarding the different nature and unique characteristics of optical and radar images, there are motivations for their fusion. This paper is aimed to identify an optimal fusion of radar and optical images to overcome their individual shortcomings and weaknesses. For this reason, panchromatic, multispectral, and radar images were first classified individually, and their strengths and weaknesses were evaluated. Different feature-level fusions of these data sets were then assessed followed by a decision-level fusion of their results. In both the feature and decision levels of integration, artificial neural networks were applied as the classifiers. Several post-processing methods using normalized different vegetation index, majority filter, and area filter were finally applied to the results. Overall accuracy of 92.8% and building detection accuracy of 89.1% confirmed the ability of the proposed fusion strategy of optical and radar images for building detection purposes.

Remote sensing is the most promising technique for providing crop maps, thanks to the development of satellite data at various temporal and spatial resolutions. Three-dimensional (3 D) convolutional neural networks (CNNs) have the potential to provide rich features that represent the spatial and temporal patterns of crops when applied to time-series images. This study presents a novel 3 D-CNN framework for classifying crops that is based on the fusion of time-series radar and optical images and also fully exploits 3 D spatial-temporal information. To extract deep convolutional maps, the proposed technique uses one separate sequence for each time-series dataset. To determine the label of each pixel, these extracted feature maps are passed to the concatenating layer and subsequently transmitted to the sequentially fully connected layers. The weights and biases are fine-tuned driving features from samples of crops in the proposed strategy. The proposed approach not only takes advantage of CNNs, i.e.,, automatic feature extraction, but also discovers discriminative feature maps in both spatial and temporal dimensions and preserves the growth dynamics of crop cycles. The overall accuracy (OA) of 91.3% and the kappa coefficient (KC) of 89.9% confirm the proposed method's potential. It is also shown that the suggested method outperforms other methods.

Convolutional neural networks (CNNs) have shown results superior to most traditional image understanding approaches in many fields, incl. crop classification from satellite time series images. However, CNNs require a large number of training samples to properly train the network. The process of collecting and labeling such samples using traditional methods can be both, time-consuming and costly. To address this issue and improve classification accuracy, generating virtual training labels (VTL) from existing ones is a promising solution. To this end, this study proposes a novel method for generating VTL based on sub-dividing the training samples of each crop using self-organizing maps (SOM), and then assigning labels to a set of unlabeled pixels based on the distance to these sub-classes. We apply the new method to crop classification from Sentinel images. A three-dimensional (3D) CNN is utilized for extracting features from the fusion of optical and radar time series. The results of the evaluation show that the proposed method is effective in generating VTL, as demonstrated by the achieved overall accuracy (OA) of 95.3% and kappa coefficient (KC) of 94.5%, compared to 91.3% and 89.9% for a solution without VTL. The results suggest that the proposed method has the potential to enhance the classification accuracy of crops using VTL.

در این مقاله، به‌منظور رفع برخی محدودیت‌های شناسایی ساختمان در تصاویر چندطیفی، از دادة SAR به‌منزلة دادة مکمل استفاده می‌شود. در روش پیشنهادی، برای استفادة هم‌زمان از اطلاعات مفید در تصاویر رادار و چندطیفی، استراتژی مبتنی‌بر تلفیق تصاویر، با هدف شناسایی ساختمان، مطرح می‌شود. همچنین، ازآن‌جاکه انتخاب ویژگی نقش بسزایی در شناسایی و طبقه‌بندی عوارض دارد، اغلب روش‌های مرسوم و رایج در این زمینه، مانند الگوریتم ژنتیک، نیازمند داده‌های آموزشی‌اند؛ اما دردسترس‌نبودن همیشگی این نوع داده‌های آموزشی یکی از دغدغه‌های مهم محققان به‌شمار می‌آید. پس در این تحقیق، دو روش انتخاب ویژگی فیلترمبنا بررسی می‌شود تا مشخص شود آیا روش‌های یادشده می‌توانند، در مواقع لازم (نبودِ دادة آموزشی)، جایگزین الگوریتم ژنتیک شوند؟ بنابراین، در پژوهش حاضر، ابتدا بردار ویژگی‌ بهینه از تصویر چندطیفی و SAR، با سه روش MNF وPCA و ژنتیک، تعیین و هریک جداگانه وارد هر دو طبقه‌بندی‌کنندة شبکة عصبی و SVM می‌شود. سپس به‌منظور رفع مشکلاتی، همچون تشابه طیفی پشت‌بام‌ها با پوشش آسفالت خیابان‌ها، در تصاویر چندطیفی و بهبود نتایج، دو تصویر چندطیفی و SAR در سطح ویژگی...