Description: Tobacco plant detection plays an important role in the management of tobacco planting. In this paper, a new algorithm based on deep neural networks is proposed to detect tobacco plants in images captured by unmanned aerial vehicles (UAVs) (called UAV images). These UAV images are characterized by a very high spatial resolution (35 $text{mm}$ ), and consequently contain an extremely high level of detail for the development of automatic detection algorithms. The proposed algorithm consists of three stages. In the first stage, a number of candidate tobacco plant regions are extracted from UAV images with the morphological operations and watershed segmentation. Each candidate region contains a tobacco plant or a nontobacco plant. In the second stage, a deep convolutional neural network is built and trained with the purpose of classifying the candidate regions as tobacco plant regions or nontobacco plant regions. In the third stage, postprocessing is performed to further remove the nontobacco plant regions. The proposed algorithm is evaluated on a UAV image dataset. The experimental results show that the proposed algorithm performs well on the detection of tobacco plants in UAV images.

Description: Multilooking is a key step in interferometric processing, especially in so-called coherent stacking interferometry approaches. In the past, multilooking algorithms were mainly implemented in the spatial domain on single interferometric pairs. With continuous development in repeat pass capabilities, multitemporal coherent synthetic aperture radar (SAR) images are now generally acquired more easily, thus providing the possibility to exploit also the temporal signature for multilooking. In this field, the possibility to carry out adaptive multilooking is fundamental for the improvement of interferometric processing. A basic similarity test has been introduced in the SqueeSAR approach, namely the Kolmogorov–Smirnov test. Furthermore, similarity tests have been discussed in terms of real-valued data vector, so only amplitude information can be utilized: The influence of the phase signal is typically ignored. To fully exploit the complex information acquired by coherent SAR systems, this paper proposes an adaptive multilooking algorithm based on complex patch (AMCP). The complex signal, which is fundamental in interferometric systems, is here exploited for the derivation of a new patch selection method. The AMCP algorithm can be applicable to all multitemporal techniques that need filtering, including the InSAR stacks of single main image and multi main images. Experiments on simulated data and real data validate that the proposed algorithm has the highlighting major advantages in improving measurement precision compared with traditional methods.

Description: Continuous monitoring of topographic heights and changes in tidal flats is challenging, as it is generally difficult to observe topographic changes from on-site measurements or remote sensing techniques with high resolution and high accuracy. In this regard, an interferometric synthetic aperture radar (In-SAR) can be an effective tool to generate precise digital elevation models (DEMs) and detect large-scale topographic changes. Nevertheless, utilizing the In-SAR to detect topographic changes in tidal flats is not practical because the average slope of tidal flats is usually less than 5°, and the overall spatial and temporal variations of height are not significant. Therefore, the accuracy of In-SAR DEMs must be high to detect meaningful topographic changes. In order to minimize the error of In-SAR DEMs, height of ambiguity and random phase deviation of interferograms should be taken into account. These two factors are related to incidence angle and baseline. We simulated topographic error levels in tidal flats for a single-pass In-SAR system such as TanDEM-X. Phase error of interferograms was derived using the relationship between In-SAR coherence and the probability density function of phase deviation. Signal-to-noise ratio and geometric decorrelation were formulated by the function of baseline, incidence angle, and surface slope. The simulation results show that the height error of the DEM was minimized to lower than 15 cm when the baseline was 1500 m with an incidence angle of 29° in the TanDEM-X system. Finally, the validation of simulation results was carried out by comparing them with TanDEM-X DEM accuracies in tidal flats.

Description: The linear operator has been widely used to detect targets of interest in multispectral/hyperspectral images, and is usually able to achieve good performance when the target is linearly separable from the background. However, when dealing with a complex scene, it is hard to find a single projection direction, along which the target can be well distinguished from all the background objects. Therefore, we propose a piecewise linear strategy (PLS) for target detection, which is based on the assumption that the desired target is generally linearly separable from each background object. PLS first divides the whole background into several partitions, and then detects the individual target for each partition by using a commonly used linear detector. Experiments with simulated and real-world multispectral/hyperspectral images show that PLS can acquire a nonlinear detection result and can outperform state-of-the-art target detection operators.

Description: Provides a listing of current staff, committee members and society officers.

Description: Presents the table of contents for this issue of the periodical.

Description: Presents the front cover for this issue of the publication.

Description: Compared with color or grayscale images, hyperspectral images deliver more informative representation of ground objects and enhance the performance of many recognition and classification applications. However, hyperspectral images are normally corrupted by various types of noises, which have a serious impact on the subsequent image processing tasks. In this paper, we propose a novel hyperspectral image denoising method based on tucker decomposition to model the nonlocal similarity across the spatial domain and global similarity along the spectral domain. In this method, 3-D full band patches extracted from a hyperspectral image are grouped to form a third-order tensor by utilizing the nonlocal similarity in a proper window size. In this way, the task of image denoising is transformed into a high-order tensor approximation problem, which can be solved by nonnegative tucker decomposition. Instead of a traditional alternative least square based tucker decomposition, we propose a hierarchical least square based nonnegative tucker decomposition method to reduce the computational cost and improve the denoising effect. In addition, an iterative denoising strategy is adopted to achieve better denoising outcome in practice. Experimental results on three datasets show that the proposed method outperforms several state-of-the-art methods and significantly enhances the quality of the corrupted hyperspectral image.

Description: Very high resolution optical remote sensing images (RSI) are often corrupted by noise. Among popular denoising methods in the state of the art, nonlocal Bayes (NLB) has led to successful results on real datasets, with high quality and reasonable computation time. However, its computation time remains prohibitive with respect to requirements of operational RSI pipelines, such as Pléiades one. In this paper, we tackle such an issue and introduce several optimizations aiming to significantly reduce the computation time required by NLB while keeping the best denoising quality (i.e., preserving edges, textures, and homogeneous areas). More precisely, our improvements consist of reducing multiple estimations of a same pixel with a masking technique and modifying the spatial extent of the similar patch search area (i.e., one of the main parts of nonlocal algorithms, such as NLB). We report several experiments and discuss optimal settings for these parameters, allowing a gain in computation time of 50% (resp. 15%) with optimized masking strategy (resp. spatial extent of the search area). When both contributions are combined, we achieve the same denoising quality as standard NLB while doubling the computation efficiency, the latter being increased fivefold if we accept a very small (lower than 0.1%) loss in quality.