Narrow Results

In order to optimize the detection capability of small and weak ground targets in remote sensing images, a remote sensing image ground weak target detection method based on multi-scale module improved Faster R-CNN is proposed. By adjusting the Region Proposal Network (RPN) and Region of Interest (RoI) pooling layers, the model is better suited for feature extraction and classification of small targets. Generate feature maps at multiple scales by introducing additional convolutional layers to process inputs at different resolutions, with each layer focusing on targets within a specific size range. These multi-scale features are integrated and input into RPN to provide richer target proposal characteristics. By using image transformation techniques such as scaling, cropping, and rotation, the training dataset is expanded to simulate different small target scenes, enhancing the model’s adaptability to changes in target size, direction, and environment. The experimental results show that the proposed method not only outperforms other models in accuracy, recall, and F1 score, but also maintains high consistency and stability across different datasets. Its significant advantage in mean accuracy (MAP) further proves its reliability and effectiveness in remote sensing applications.

Deep learning has enabled significant advancements in the classification of remote sensing images; however, the task of classifying images in remote sensing remains a formidable challenge because of the high item diversity and complexity that result from spatial and temporal combination and connection. The problem of insufficient differentiation of feature representations generated by deep learning remains, which is mostly due to the similarity and variety of inter-class and intra-class images, respectively. This paper introduces a novel hexagonal network architecture called DenseNet-169, which is based on end-to-end convolutional methods (Bi-LSTM and RNN model) known as CRLSTM-Hexnet. The proposed model comprises three distinct components: (1) a module for extracting features, (2) a feature selection module utilizing the Harris Hawk optimization (HHO) algorithm, and (3) a sub-network based on LSTM and RNN, incorporating a class attention module learning layer. Positive quantitative and qualitative findings from experiments on the RSI-CB256 multi-label dataset confirm the efficacy of our model.

Due to their propensity for stripe noise distortions, infrared remote sensing images present substantial difficulty for interpretation. Our ability to address this issue by offering an easy, efficient, and fast technique for image stripe noise correction is what makes our work unique. Our proposed solution tackles stripe noise by subtracting the mean value along the stripes from the noisy image. Additionally, we leverage the wavelet transform on the average signal to exploit the inherent sparsity of noise in the wavelet domain. This approach not only enhances denoising performance without introducing blurring effects but also enables us to recover image details with remarkable precision, all without the need for intricate algorithms, iterative processes, or training models. To validate the effectiveness of our approach, we conducted evaluations using a dataset of real-world infrared remote sensing images. This dataset encompasses a wide range of examples, featuring both real and artificially induced noise scenarios.

Discrimination between hazardous materials in the environment and ambient constituents is a fundamental problem in environmental sensing. The ubiquity of naturally occurring bacteria, plant pollen, fungi, and other airborne materials makes the task of sensing for biological warfare (BW) agents particularly challenging. The spectroscopic properties of the chemical warfare (CW) agents in the long wavelength infrared (LWIR) region are important physical properties that have been successfully exploited for environmental sensing. However, in the case of BW agents, the LWIR region affords less distinction between hazardous and ambient materials. Recent studies of the THz spectroscopic properties of biological agent simulants, particularly bacterial spores, have yielded interesting and potentially useful spectral signatures of these materials. It is anticipated that with the advent of new THz sources and detectors, a novel environmental sensor could be designed that exploits the peculiar spectral properties of the biological materials. We will present data on the molecular spectroscopy of several CW agents and simulants as well as some THz spectroscopy of the BW agent simulants that we have studied to date, and discuss the prospectus with regard to detection probabilities through the application of sensor system modeling.

An innovative passive standoff system for the detection of chemical/biological agents is described. The spectral, temporal and spatial resolution of the data collected are all adjustable in real time, making it possible to keep the tradeoff between the sensor operating parameters at optimum at all times. The instrument contains no macro-scale moving parts and is therefore an excellent candidate for the development of a robust, compact, lightweight and low-power-consumption sensor. The design can also serve as a basis for a wide variety of spectral instruments operating in the visible, NIR, MWIR, and LWIR to be used for surveillance, process control, and biomedical applications.

To address the problem of sources and sinks of atmospheric CO₂, measurements are needed on a global scale. Satellite instruments show promise, but typically measure the total column. Since sources and sinks at the surface represent a small perturbation to the total column, a precision of better than 1% is required. No species has ever been measured from space at this level. Over the last three years, we have developed a small instrument based upon a Fabry-Perot interferometer that is highly sensitive to atmospheric CO₂. We have tested this instrument in a ground based configuration and from aircraft platforms simulating operation from a satellite. The instrument is characterized by high signal to noise ratio, fast response and great specificity. We have performed simulations and instrument designs for systems to detect, H₂O, CO, ¹³CO₂, CH₄, CH₂O, NH₃, SO₂, N₂O, NO₂, and O₃. The high resolution and throughput, and small size of this instrument make it adaptable to many other atmospheric species. We present results and discuss ways this instrument can be used for ground, aircraft or space based surveillance and the detection of pollutants, toxics and industrial effluents in a variety of scenarios including battlefields, industrial monitoring, or pollution transport.

Heavy loads of aerosols in the air have considerable health effects in individuals who suffer from chronic breathing difficulties. This problem is more acute in the Middle-East, where dust storms in winter and spring transverse from the neighboring deserts into dense populated areas. Discrimination between the dust types and association with their source can assist in assessment of the expected health effects. A method is introduced to characterize the properties of dense dust clouds with passive IR spectral measurements. First, we introduce a model based on the solution of the appropriate radiative transfer equations. Model predictions are presented and discussed. Actual field measurements of silicone-oil aerosol clouds with an IR spectro-radiometer are analyzed and compared with the theoretical model predictions. Silicone-oil aerosol clouds have been used instead of dust in our research, since they are composed of one compound in the form of spherical droplets and their release is easily controlled and repetitive. Both the theoretical model and the experimental results clearly show that discrimination between different dust types using IR spectral measurements is feasible. The dependence of this technique on measurement conditions, its limitations, and the future work needed for its practical application of this technique is discussed.

We have developed a hyperspectral deconvolution algorithm that sharpens the spectral dimension in addition to the more usual across-track and along-track dimensions. Using an individual three-dimensional model for each pixel's point spread function, the algorithm iteratively applies maximum likelihood criteria to reveal previously hidden features in the spatial and spectral dimensions. Of necessity, our solution is adaptive to unreported across-track and along-track vibrations with amplitudes smaller than the ground sampling distance. We sense and correct these vibrations using a combination of maximum likelihood deconvolution and gradient descent registration that maximizes statistical correlations over many bands. Test panels in real hyperspectral imagery show significant improvement when locations are corrected. Tests on simulated imagery show that the precision of relative corrected positions improves by about a factor of two.

A cross-comparison method was used to assess the SPOT-6 optical satellite imagery against Chinese GF-1 imagery using three types of indicators: spectral and color quality, fusion effect and identification potential. More specifically, spectral response function (SRF) curves were used to compare the two imagery, showing that the SRF curve shape of SPOT-6 is more like a rectangle compared to GF-1 in blue, green, red and near-infrared bands. NNDiffuse image fusion algorithm was used to evaluate the capability of information conservation in comparison with wavelet transform (WT) and principal component (PC) algorithms. The results show that NNDiffuse fused image has extremely similar entropy vales than original image (1.849 versus 1.852) and better color quality. In addition, the object-oriented classification toolset (ENVI EX) was used to identify greenlands for comparing the effect of self-fusion image of SPOT-6 and inter-fusion image between SPOT-6 and GF-1 based on the NNDiffuse algorithm. The overall accuracy is 97.27% and 76.88%, respectively, showing that self-fused image of SPOT-6 has better identification capability.

Classification of land cover based on hyperspectral data is very challenging because typically tens of classes with uneven priors are involved, the inputs are high dimensional, and there is often scarcity of labeled data. Several researchers have observed that it is often preferable to decompose a multiclass problem into multiple two-class problems, solve each such subproblem using a suitable binary classifier, and then combine the outputs of this collection of classifiers in a suitable manner to obtain the answer to the original multiclass problem. This approach is taken by the popular error correcting output codes (ECOC) technique, as well by the binary hierarchical classifier (BHC). Classical techniques for dealing with small sample sizes include regularization of covariance matrices and feature reduction. In this paper we address the twin problems of small sample sizes and multiclass settings by proposing a feature reduction scheme that adaptively adjusts to the amount of labeled data available. This scheme can be used in conjunction with ECOC and the BHC, as well as other approaches such as round-robin classification that decompose a multiclass problem into a number of two (meta)-class problems. In particular, we develop the best-basis binary hierarchical classifier (BB-BHC) and best basis ECOC (BB-ECOC) families of models that are adapted to "small sample size" situations. Currently, there are few studies that compare the efficacy of different approaches to multiclass problems in general settings as well as in the specific context of small sample sizes. Our experiments on two sets of remote sensing data show that both BB-BHC and BB-ECOC methods are superior to their nonadaptive versions when faced with limited data, with the BB-BHC showing a slight edge in terms of classification accuracy as well as interpretability.

An image clustering method that is based on the particle swarm optimizer (PSO) is developed in this paper. The algorithm finds the centroids of a user specified number of clusters, where each cluster groups together with similar image primitives. To illustrate its wide applicability, the proposed image classifier has been applied to synthetic, MRI and satellite images. Experimental results show that the PSO image classifier performs better than state-of-the-art image classifiers (namely, K-means, Fuzzy C-means, K-Harmonic means and Genetic Algorithms) in all measured criteria. The influence of different values of PSO control parameters on performance is also illustrated.

This paper proposes boundary parallel-like index (BPI) to describe shape features for high-resolution remote sensing image classification. Parallel-like boundary is found to be a discriminating clue which can reveal the shape regularity of segmented objects. Therefore, multi-orientation distance projections were constructed to measure and quantify parallel-like information. The discriminating ability was tested using original and segmented ground objects, respectively. The proposed BPI showed better discrimination for both original and segmented data than for other shape features, especially for buildings. This was also confirmed by the considerably higher accuracy of BPI in building classification experiments of high-resolution remote sensing imagery. It suggests the proposed BPI is useful for building related applications.

Current state-of-the-art detectors achieved impressive performance in detection accuracy with the use of deep learning. However, most of such detectors cannot detect objects in real time due to heavy computational cost, which limits their wide application. Although some one-stage detectors are designed to accelerate the detection speed, it is still not satisfied for task in high-resolution remote sensing images. To address this problem, a lightweight one-stage approach based on YOLOv3 is proposed in this paper, which is named Squeeze-and-Excitation YOLOv3 (SE-YOLOv3). The proposed algorithm maintains high efficiency and effectiveness simultaneously. With an aim to reduce the number of parameters and increase the ability of feature description, two customized modules, lightweight feature extraction and attention-aware feature augmentation, are embedded by utilizing global information and suppressing redundancy features, respectively. To meet the scale invariance, a spatial pyramid pooling method is used to aggregate local features. The evaluation experiments on two remote sensing image data sets, DOTA and NWPU VHR-10, reveal that the proposed approach achieves more competitive detection effect with less computational consumption.

Remote sensing techniques have been developed over the past decades to acquire data without being in contact of the target object or data source. Their application on land-cover image segmentation has attracted significant attention during recent years. With the help of satellites, scientists and researchers can collect and store high-resolution image data that can be further be processed, segmented, and classified. However, these research results have not yet been synthesized to provide coherent guidance on the effect of variant land-cover segmentation processes. In this paper, we present a novel model that augments segmentation using smaller networks to segment individual classes. The combined network is trained on the same data but with the masks, combined and trained using categorical cross entropy. Experimental results show that the proposed method produces the highest mean IoU (Intersection of Union) as compared against several existing state-of-the-art models on the DeepGlobe dataset.

Image time series, such as Satellite Image Time Series (SITS) or MRI functional sequences in the medical domain, carry both spatial and temporal information. In many pattern recognition applications such as image classification, taking into account such rich information may be crucial and discriminative during the decision making stage. However, the extraction of spatio-temporal features from image time series is difficult to handle due to the complex representation of the data cube. In this paper, we present a strategy based on Random Walk to build a novel segment-based representation of the data, passing from a 2D$+t$ dimension to a 2D one, more easily manipulable and without losing too much spatial information. Such new representation is then used to feed a classical Convolutional Neural Network (CNN) in order to learn spatio-temporal features with only 2D convolutions and to classify image time series data for a particular classification problem. The influence of the way the 2D$+t$ data are represented, as well as the impact of the network architectures on the results, are carefully studied. The interest of this approach is highlighted on a remote sensing application for the classification of complex agricultural crops.

In recent years, with the development of remote sensing technology and the enhancement of the value of remote sensing images in military and civil fields, remote sensing image object segmentation has also received more and more attention. This paper mainly studies the application of instance segmentation based on deep convolutional neural network in the remote sensing image. This paper proposes an attention balanced feature pyramid module, which strengthens multi-level features and uses the attention module to suppress the interference features of noise in the complex background. In addiction, Soft-NMS is introduced to improve the performance of the network, and GIoU loss is introduced to improve the effect of object detection. The proposed network improves the average detection and segmentation accuracy (mAP) values from $41.75\%$ and $35.34\%$ to $43.05\%$ and $36.02\%$, respectively.

Expert systems and image understanding have traditionally been considered as two separate application fields of artificial intelligence (AI). In this paper it is shown, however, that the idea of building an expert system for image understanding may be fruitful. Although this paper may serve as a framework for situating existing works on knowledge-based vision, it is not a review paper. The interested reader will therefore be referred to some recommended survey papers in the literature.

In remote sensing the intensities from a multispectral image are used in a classification scheme to distinguish different ground cover from each other. An example is given where different soil types are classified. A digitized complete scene from a satellite sensor consists of a large amount of data and in future image sensors the resolution and the number of spectral bands will increase even further. Data parallel computers are therefore well-suited for these types of classification algorithms. This article will focus on three supervised classified algorithms: the Maximum Likelihood, the K-Nearest Neighbor and the Backpropagation algorithm, together with their parallel implementations. They are implemented on the Connection Machine/200 in the high-level language C*. The algorithms are finally tested and compared on an image registered over western Estonia.

Recently, a kind of structured neural networks (SNNs) explicitly devoted to multisensor image recognition and aimed at allowing the interpretation of the "network behavior" was presented in Ref. 1. Experiments reported in Ref. 1 pointed out that SNNs provide a trade-off between recognition accuracy and interpretation of the network behavior. In this paper, the combination of multiple SNNs, each of which has been trained on the same data set, is proposed as a means to improve recognition results, while keeping the possibility of interpreting the network behavior. A simple method for interpreting the "collective behaviors" of such SNN ensembles is described. Such an interpretation method can be used to understand the different kinds of "solutions" learned by the SNNs belonging to an ensemble. In addition, as compared with the interpretation method presented in Ref. 1, it is shown that the knowledge embodied in an SNN can be translated into a set of understandable "recognition rules". Experimental results on the recognition of multisensor remote-sensing images (optical and radar images) are reported in terms of both recognition accuracy and network-behavior interpretation. An additional experiment on a multisource remote-sensing data set is described to show that SNNs can also be effectively used for multisource recognition tasks.

This paper proposes a method for remote sensing based land cover/land use classification of urban areas. The method consists of the following four main stages: feature extraction, feature coding, feature selection and classification. In the feature extraction stage, statistical, textural and Gabor features are computed within local image windows of different sizes and orientations to provide a wide variety of potential features for the classification. Then the features are encoded and normalized by means of the Self-Organizing Map algorithm. For feature selection a CART (Classification and Regression Trees) based algorithm was developed to select a subset of features for each class within the classification scheme at hand. The selected subset of features is not attached to any specific classifier. Any classifier capable of representing possible skewed and multi-modal feature distributions can be employed, such as multi-layer perceptron (MLP) or k-nearest neighbor (k-NN). The paper reports experiments in land cover/land use classification with the Landsat TM and ERS-1 SAR images gathered over the city of Lisbon to show the potentials of the proposed method.