Accurate agricultural parcel boundary delineation is essential in remote sensing applications, yet traditional supervised methods require extensive annotated datasets and often fail to generalize across diverse landscapes. The Segment Anything Model (SAM), a foundational model for zero-shot segmentation, provides scalability but struggles with certain remote sensing challenges, particularly agricultural parcels.In this paper, we propose a novel approach to enhance SAM’s performance by leveraging its embeddings to extract meaningful features. Our method applies principal component analysis (PCA) for dimensionality reduction, high-frequency decomposition, and guided filtering to enhance input images, aligning them better with SAM’s strengths. By refining the input data through these steps, we improve SAM’s ability to delineate parcel boundaries effectively. Experimental results demonstrate consistent improvements across SAM back-bone sizes and parameter settings, achieving higher accuracy in segmentation metrics such as under-segmentation (US) rate, over-segmentation (OS) rate, intersection over union (IoU), and false negative (FN) rate.

This paper explores the application of a singular expanding path of Frequency Attention U-Net (FAUNet), specifically employing its frequency attention mechanism for road detection in remote sensing. Contrasting with the full dual-path architecture of the recently proposed FAUNet, this study cap-italizes on only the high-frequency attentive path, tailored for edge detection in road segmentation tasks. By focusing on this single path, the modified FAUNet is adept at highlighting the intricate details necessary for accurate road boundary identification in high resolution remote sensing images. Comparative evaluations are conducted against traditional models like U-Net, U-Net++, and a generic CNN under consistent experimental conditions, including identical datasets, loss functions, and training loops. The findings reveal that this adapted version of FAUNet outperforms the other models in key performance metrics such as recall, F1 score, and Intersection over Union (IoU)

Since clutter encountered in Ground Penetrating Radar (GPR) systems deteriorates the performance of target detection algorithms, clutter removal is an active research area in the GPR community. In this paper, instead of Convolutional Neural Network (CNN) architectures used in the recently proposed deep learning-based clutter removal methods, we introduce Declutter Vision Transformers (DC-ViT) to remove the clutter. Transformer Encoders in DC-ViT provide an alternative to CNNs which has limitations to capture long-range dependencies due to its local operations. Also, the implementation of a convolutional layer instead of Multilayer Perceptron (MLP) in the Transformer Encoder increases the capturing ability of local dependencies. While deep features are extracted with blocks consisting of Transformer encoders arranged sequentially, losses during information flow are reduced by using dense connections between these blocks. Our proposed DC-ViT was compared with Low-Rank and Sparse methods such as Robust Principle Component Analysis (RPCA), Robust Nonnegative Matrix Factorization (RNMF), and CNN-based deep networks such as Convolutional Auto-Encoder (CAE) and CR-NET. In comparisons made with the hybrid dataset, DC-ViT is 2.5% better in PSNR results than its closest competitor. As a result of the tests we conducted using our experimental GPR data, the proposed model provided an improvement of up to 20%, compared to its closest competitor in terms of SCR.

Speckle noise in SAR images distorts the image of the target and its surroundings, making difficult the target recognition task. Therefore, decomposition process of the speckle noise from the SAR images is important for radar automatic target recognition applications. Besides since the succes of the deep networks depends on the amount of data used in the training stage data augmentation increases classification rates. In this study, a new data augmentation approach based on despeckling has been proposed rather than the classical data augmentation techniques used in the processing of natural images in order to increase the deep learning-based radar target classification performance. Edge Avoiding Wavelet filter is used for speckle reduction task. Classification performances for original, despeckled and despeckling based data augmented datasets are compared on two traditional and basic CNN models. The experimental results show that despeckling based data augmentation method can improve the deep learning based radar automatic target recognition classification performance.

Object detection in high resolution satellite images has recently become a major concern in new geospatial information methods. The higher spatial resolution with spectral information provides better detection results. Therefore, increasing the image resolution prior to the object detection is important. For this purpose, pansharpening, which uses complementary information from MS and PAN images, is gaining popularity as it helps to increase spatial resolution while preserving the spectral information. This study proposes a detailed enhanced scheme for pansharpening to improve the detection results. Several deep learning models are trained on raw dataset, as well as on the detail enhanced pansharpened images. It is shown that the training stage using proposed detail enhanced scheme provides better detection results compared to classical pansharpening or raw data based training for different deep networks.

In this paper a new infrared and visible image fusion (IVIF) method which combines the advantages of optimization and deep learning based methods is proposed. This model takes the iterative solution used by the alternating direction method of the multiplier (ADMM) optimization method, and uses algorithm unrolling to obtain a high performance and efficient algorithm. Compared with traditional optimization methods, this model generates fusion with 99.6% improvement in terms of image fusion time, and compared with deep learning based algorithms, this model generates detailed fusion images with 99.1% improvement in terms of training time. Compared with the other state-of-the-art unrolling based methods, this model performs 26.7% better on average in terms of Average Gradient (AG), Cross Entropy (CE), Mutual Information (MI), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Loss (SSIM) metrics with a minimal testing time cost.

The clutter encountered in ground-penetrating radar (GPR) systems seriously affects the performance of the subsurface target detection methods. A new clutter removal method based on convolutional autoencoders (CAEs) is introduced. The raw GPR image is encoded via successive convolution and pooling layers and then decoded to provide the clutter-free GPR image. The loss function is defined in terms of the reference clutter-free target image and the decoder output is optimized to learn the weight coefficients from the raw data. The method is compared to the conventional subspace methods, recently proposed nonnegative matrix factorization, as well as low-rank and sparse decomposition (LRSD) methods and dictionary separation-based morphological component analysis. CAE and its deeper version deep CAE (DCAE) are trained by several scenarios generated by the electromagnetic simulation tool gprMax. Simulation results demonstrate the effectiveness of the proposed method for challenging scenarios. While for real GPR image, the simulated data trained networks remain slightly behind the LRSD methods for the dry case, nonetheless, they outperform the aforementioned processing techniques for the more challenging wet case.

An efficient image enhancement method should improve the contrast in the image while keeping the edge and color information. Since existing approaches seem not to fulfill all these demands, a hybrid approach which will combine advantages of individual approaches is proposed in this work. The multiscale bilateral filter is replaced by an iterative joint version where the output is used as guidance image for the next iterations. Then a multiscale structure is designed by the appropriate modifications of the spatial and range kernels as in the multiscale bilateral filter. A final refining is performed by the local use of the Hazy Image Model based method (HIM) on the resulting image. Visual and quantitative comparisons with conventional Discrete Wavelet Transform and Singular Value Decomposition based method (DWT-SVD), Regularized Histogram Equalization with Discrete Cosine Transform method (RHE-DCT), Bilateral Filtering based method (BF), and HIM method demonstrate the superiority of the proposed method and the resulting hybrid method for remote sensing image enhancement.

The clutter encounters in Ground Penetrating Radar (GPR) systems decrease the performance of target detection methods. This work presents a clutter aware detection method using deep learning. The clutter is learned and eliminated prior to the detection by a low rank and sparse decomposition of the raw data matrix. The deep networks are fed with clutter free data with increased target visibility. GPR scenarios are generated by gprMax. Recently proposed robust non-negative matrix factorization (RNMF) with less complexity and better visual performance among low rank and sparse decomposition (LRSD) methods, performs the clutter removal. Besides the traditional Faster R -CNN, Yolo5 and EfficientDet are used in the detection step. Results validate that using clutter removed data increases the detection rate of deep networks.

The clutter encountered in the ground-penetrating radar (GPR) system severely decreases the visibility of subsurface objects, thus highly degrading the performance of the target detection algorithms. This letter presents a new clutter removal method based on nonnegative matrix factorization (NMF). The raw GPR data are represented as the sum of low-rank and sparse matrices, which correspond to the clutter and target components, respectively. The low-rank and sparse decomposition is performed using a robust version of NMF called RNMF. Although similar to the robust principal component analysis (PCA) (RPCA), which is recently widely used in image processing applications as well as in GPR, the proposed method is faster and has enhanced results. The state-of-the-art clutter removal methods, morphological component analysis (MCA), RPCA, besides the conventional PCA, have been included for comparison for both simulated and real data sets. The visual and quantitative results demonstrate that the proposed RNMF method outperforms the others. Moreover, it is 25 times faster than the RPCA for the given regularization parameter values.

Ground-penetrating radar (GPR) is a popular technique to detect buried objects such as landmines. It is well known that the target detection process in GPR is highly affected by clutter since targets are buried at shallow depths. Thus, design a target detection scheme with less false alarm rate is the ultimate goal of a GPR system. Therefore, various subspace or multiscale methods has been proposed for clutter removal in GPR images. In this paper, multiscale directional bilateral filter (MDBF) is integrated in GPR scheme. Due to its range and spatial parameters determined in order to enhance a predefined metric appropriate to the desired application, MDBF can provide a more flexible decomposition unlike the other multiresolution approaches with constants kernels. This property enables the input images to be decomposed into the directional detail subbands with different geometrical structures. Then, the informative subbands for target are preserved and the inverse MDBF is applied to reconstruct noise-free image. The results show the superiority of our algorithm compared to other multiresolution based methods proposed in the literature.

Ground Penetrating Radar (GPR) is one of the most popular subsurface sensing devices and has a wide range of applications, e.g., buried object detection. In this study, Least Mean Square (LMS) approach is used to solve buried object detection problem. Point of interest located in each depth location of 2D GPR signal is estimated from previous samples by using separate 1D LMS algorithms and prediction errors defined as the difference between the measured and estimated values are aggregated. If calculated error exceeded a predefined threshold, it is decided that a buried object exists at that location. The proposed approach is tested with a realistic data set simulated by using a new version of gprMax electromagnetic modeling software. The data set consists of several different soil types, objects, different burial depths and surface types. Resulting Receiver Operating Characteristic (ROC) curves demonstrate the performance of the proposed method.

Ground-penetrating radar (GPR) is one of the most popular subsurface sensing devices and has a wide range of applications, e.g., target detection. It is well known that the target detection process in the GPR is highly affected by clutter. Especially, in the case of landmine detection, since targets are located near the surface, a target signal may be completely covered by the clutter. Thus, clutter reduction must be performed prior to any target detection scheme in the GPR. Singular value decomposition, principal component analysis, and independent component analysis are commonly used for clutter removal. They all aim to decompose the GPR images into subcomponents that represent the clutter and the target separately. In this letter, we propose a sparse model for differentiating the target and the clutter using appropriate dictionaries based on morphological component analysis (MCA). Calculated sparse coefficients and corresponding dictionaries are used to reconstruct the clutter and the target components. Visual and quantitative results validate that the proposed MCA-based method has higher performance than the state-of-the-art clutter reduction methods.

An efficient pansharpening method should inject the missing geometric information to the multispectral (MS) image while preserving its radiometric information. Widely used additive wavelet transform-based pansharpening methods extract the missing high-frequency information by decomposing the panchromatic (PAN) image and adding the detail layers to the low-resolution MS (LRM) image. However, this approach causes a redundant detail injection, leading to artifacts in the fusion result. In this letter, we propose to decompose the high-resolution-PAN image using an edge-preserving decomposition which will decrease the amount of redundant high-frequency injection. The missing high-frequency information of the LRM image is obtained by the decomposition of the PAN image using a multiscale bilateral filter. The spatial and range parameters of the bilateral filter are optimized so as to enhance spatial and spectral metrics. The fusion results are compared with the widely used additive wavelet luminance proportional (AWLP) and recently proposed improved AWLP fusion methods. The resulting images as well as evaluation metrics demonstrate that the proposed injection approach has better performance.

A new fusion method based on bilateral pyramid for multispectral and panchromatic images is presented. The fused image is obtained by two different rules: substitutive and additive methods. Bilateral pyramid is a multiscale decomposition method which decomposes an input image into a base layer representing the low frequency content and several detail layers representing the high frequency part of the image. In substitutive method, both MS and PAN images are decomposed using bilateral pyramid. The detail layers of the PAN image are added to the base layer of the MS image. In additive method, the detail layers of the PAN image are directly added to the MS image. The proposed method is compared with the widely used IHS (intensity-hue-saturation), ATWT substitutive and ATWT additive fusion methods. The resulting images as well as evaluation metrics demonstrate that the proposed algorithm has better performance.

A new speckle reduction algorithm based on lattice filters for SAR imaging is presented. In the new method, the subband decomposition of the speckled image is performed using lattice filters. The noisy image is decomposed into subband images using high-pass and low-pass filters having lattice structure, then a threshold value is estimated according to noise variance in each subband and soft-thresholding is applied on the subband images. The despeckled image is obtained from the thresholded subband images using the inverse lattice filter. The proposed speckle reduction method is applied to RADARSAT/SAR images. The performance of the proposed method has also been compared with median filtering, and discrete and stationary wavelet transform based speckle reduction methods. Results show that the proposed method may be used efficiently for speckle noise reduction in SAR images.

https://doi.org/10.1109/igarss.2005.1526650

A new data extrapolation technique which utilizes pole extraction based GPOF method is proposed to fit the scattering characteristics of an object. The proposed data extrapolation method is very efficient for data extrapolation when applied to the ISAR data. Modeling the data as the superposition of complex exponential signals, GPOF and Prony methods do not guarantee a stable prediction filter like the MCM method, while Burg ensures. Meanwhile, the performance of GPOF decreases with the increasing number of scattering centers but the radar images obtained using this extrapolated data is still more accurate than those obtained using the extrapolated data generated by other methods.

The reconstruction algorithms based on microwave diffraction tomography have been developed to visualize multidimensional cross-section of an object. A wavelet based approach is discussed for the solution of microwave diffraction tomography problem. The matrix equation is described in the wavelet domain. The major advantages of this solution is that the wavelet transform can usefully be severely truncated, that is, turned into sparse expansions. The use of sparse matrix techniques leads to considerable reduction in computation time and data.