In this paper, we evaluate the generalization power of deep features (ConvNets) in two new scenarios: aerial and remote sensing image classification. We experimentally ConvNets trained for recognizing everyday objects classification images. obtained best results images, while sensing, they performed well but were outperformed by low-level color descriptors, such as BIC. also present a correlation analysis, showing potential combining/fusing different with other descriptors or even combining...

Flooding is the world's most costly type of natural disaster in terms both economic losses and human causalities. A first essential procedure towards flood monitoring based on identifying area vulnerable to flooding, which gives authorities relevant regions focus. In this work, we propose several methods perform flooding identification high-resolution remote sensing images using deep learning. Specifically, some proposed techniques are upon unique networks, such as dilated deconvolutional...

Semantic segmentation requires methods capable of learning high-level features while dealing with large volume data. Toward such goal, convolutional networks can learn specific and adaptable based on the However, these are not processing a whole remote sensing image, given its huge size. To overcome limitation, image is processed using fixed size patches. The definition input patch usually performed empirically (evaluating several sizes) or imposed (by network constraint). Both strategies...

Urban forests are an important part of any city, given that they provide several environmental benefits, such as improving urban drainage, climate regulation, public health, biodiversity, and others. However, tree detection in cities is challenging, the irregular shape, size, occlusion, complexity areas. With advance technologies, deep learning segmentation mapping methods can map accurately. We applied a region-based CNN object instance algorithm for semantic canopies environments based on...

The performance of image classification is highly dependent on the quality extracted features. Concerning high resolution remote images, encoding spatial features in an efficient and robust fashion key to generating discriminatory models classify them. Even though many visual descriptors have been proposed or successfully used encode sensing some applications, using this sort demand more specific description techniques. Deep Learning, emergent machine learning approach based neural networks,...

Urban forests contribute to maintaining livability and increase the resilience of cities in face population growth climate change. Information about geographical distribution individual trees is essential for proper management these systems. RGB high-resolution aerial images have emerged as a cheap efficient source data, although detecting mapping single an urban environment challenging task. Thus, we propose evaluation novel methods tree crown detection, most not been investigated remote...

Land cover classification is a task that requires methods capable of learning high-level features while dealing with high volume data. Overcoming these challenges, Convolutional Networks (ConvNets) can learn specific and adaptable depending on the data while, at same time, classifiers. In this work, we propose novel technique to automatically perform pixel-wise land classification. To best our knowledge, there no other work in literature semantic segmentation based data-driven feature...

Over the past decade, Convolutional Networks (ConvNets) have renewed perspectives of research and industrial communities. Although this deep learning technique may be composed multiple layers, its core operation is convolution, an important linear filtering process. Easy fast to implement, convolutions actually play a major role, not only in ConvNets, but digital image processing analysis as whole, being effective for several tasks. However, aside from convolutions, researchers also proposed...

Classical and more recently deep computer vision methods are optimized for visible spectrum images, commonly encoded in grayscale or RGB colorspaces acquired from smartphones cameras. A uncommon source of images exploited the remote sensing field satellite aerial images. However development pattern recognition approaches these data is relatively recent, mainly due to limited availability this type as until they were used exclusively military purposes. Access imagery, including spectral...

Recent developments and research in modern machine learning have led to substantial improvements the geospatial field.Although numerous deep architectures models been proposed, majority of them solely developed on benchmark datasets that lack strong real-world relevance.Furthermore, performance many methods has already saturated these datasets.We argue a shift from model-centric view complementary data-centric perspective is necessary for further accuracy, generalization ability, real impact...

It is undeniable that aerial/satellite images can provide useful information for a large variety of tasks. But, since these are always taken from above, some applications benefit complementary provided by other perspective views the scene, such as ground-level images. Despite number public repositories both georeferenced photographs and aerial images, there lack benchmark datasets allow development approaches exploit benefits complementarity aerial/ground imagery. In this article, we present...

Multi-task learning has proven to be effective in improving the performance of correlated tasks. Most existing methods use a backbone extract initial features with independent branches for each task, and exchange information between usually occurs through concatenation or sum feature maps branches. However, this type does not directly consider local characteristics image nor level importance correlation In paper, we propose semantic segmentation method, MTLSegFormer, which combines...

In this paper, we analyse the use of convolutional neural networks (CNNs or ConvNets) to discriminate vegetation species with few labelled samples. To best our knowledge, is first work dedicated investigation deep features in such task. The experimental evaluation demonstrate that significantly outperform wellknown feature extraction techniques. achieved results also show it possible learn and classify patterns even This makes approach feasible for real-world mapping applications, where...

Geographic mapping of coffee crops by using remote sensing images and supervised classification has been a challenging research subject. Besides the intrinsic problems caused nature multi-spectral information, are non-seasonal usually planted in mountains, which requires encoding learning huge diversity patterns during classifier training. In this paper, we propose new approach for automatic combining two recent trends on pattern recognition applications: deep fusion/selection features from...

Plant phenology studies rely on long-term monitoring of life cycles plants. High-resolution unmanned aerial vehicles (UAVs) and near-surface technologies have been used for plant monitoring, demanding the creation methods capable locating, identifying species through time space. However, this is a challenging task given high volume data, constant data missing from temporal dataset, heterogeneity profiles, variety visual patterns, unclear definition individuals' boundaries in communities. In...

Vegetation segmentation in high resolution images acquired by unmanned aerial vehicles (UAVs) is a challenging task that requires methods capable of learning high-level features while dealing with fine-grained data. In this paper, we propose combination different semantic based on Convolutional Networks (ConvNets) to obtain highly accurate individuals vegetation species. The objective not only learn specific and adaptable depending the data, but also combine appropriate classifiers. We...

The identification of bridges in major infrastructure works is crucial to provide information about the status these constructions and support possible decision-making processes. Typically, this performed by human agents that must detect into large-scale datasets, analyzing image image, a time-consuming task. In paper, we propose novel tool perform bridge detection remote sensing datasets. This implements deep learning-based algorithm, Faster R-CNN (Regions with CNN features), technique...