It is challenging to accurately detect camouflaged objects from their highly similar surroundings. Existing methods mainly leverage a single-stage detection fashion, while neglecting small objects with low-resolution fine edges requires more operations than the larger ones. To tackle camouflaged object detection (COD), we are inspired by humans attention coupled with the coarse-to-fine detection strategy, and thereby propose an iterative refinement framework, coined SegMaR, which integrates Segment, Magnify and Reiterate in a multi-stage detection fashion. Specifically, we design a new discriminative mask which makes the model attend on the fixation and edge regions. In addition, we leverage an attention-based sampler to magnify the object region progressively with no need of enlarging the image size. Extensive experiments show our SegMaR achieves remarkable and consistent improvements over other state-of-the-art methods. Especially, we surpass two competitive methods 7.4% and 20.0% respectively in average over standard evaluation metrics on small camouflaged objects. Additional studies provide more promising insights into Seg-MaR, including its effectiveness on the discriminative mask and its generalization to other network architectures. Code is available at https://github.com/dlut-dimt/SegMaR.

Segment, Magnify and Reiterate: Detecting Camouflaged Objects the Hard Way

Received: 18 July 2020 Accepted: 20 December 2020 Background/objectives: The objective of the study is to increase the resolution and radiate sharper beam towards the user in mobile communication using Smart Antenna. Methodology: The Conventional and Subspace Algorithms from the literature are studied and simulated in MATLAB so that the foundation is laid for better detection of algorithms and radiation formation. The results are explained for varying number of antenna elements and mobiles sources placed close to far. Findings: The classical direction of arrival algorithms namely CAPCON, Maximum Entropy Method, Maximum Likelihood Method are used to find the direction of mobile users based on the computation of the power spectrum. Several methods namely Least Mean Square, Griffiths Method, Variable Step Size Griffiths and Recursive Least Square are used to form the main beam for the user detected by the direction of arrival algorithms. Novelty/improvements: In order to take further the research on enhancing the resolution and having a higher convergence rate with reasonable step size, this paper presents the well-known conventional and modern algorithms from the literature. The results are simulated are well described for performing parameters.

Conventional and Subspace Algorithms for Mobile Source Detection and Radiation Formation

In this work, we apply deep learning to estimate the direction-of-arrival (DoA) of multiple narrowband signals with a uniform linear array in a coherent environment. First, the logarithmic eigenvalue-based classification network (LogECNet) is introduced to enhance signal number detection accuracy in challenging scenarios, such as the low signal-to-noise (SNR) regime and limited snapshots. Next, a multi-label classification model called the root-spectrum network (RSNet) is devised to estimate the DoAs using the signal number inferred by LogECNet. In the proposed architecture, the full-row Toeplitz matrices reconstruction (FTMR), which exploits all rows of the signal covariance matrix (SCM), is combined with LogECNet and RSNet to inversely map the SCM to the numerical DoAs in the coherent scenario. It is shown that the eigenvalues factorized from the FTMR output matrix become more robust sources for signal enumeration than those of the forward/backward spatial smoothing (FBSS) algorithm. Furthermore, the logarithmic scaling of the eigenvalues of the FTMR results in LogECNet outperforming other detectors. The simulation results show our proposed method not only improves the signal number detection and angular estimation performance, but also achieves the complexity reduction with respect to the prior schemes.

Deep Learning-Aided Coherent Direction-of-Arrival Estimation With the FTMR Algorithm

This article proposes a novel active localization method based on the mixed polarization multiple signal classification (MP-MUSIC) algorithm for positioning a metal target or an insulator target in the underwater environment by using a uniform circular antenna (UCA). The boundary element method (BEM) is introduced to analyze the boundary of the target by use of a matrix equation. In this method, an electric dipole source as a part of the locating system is set perpendicularly to the plane of the UCA. As a result, the UCA can only receive the induction field of the target. The potential of each electrode of the UCA is used as spatial-temporal localization data, and it does not need to obtain the field component in each direction compared with the conventional fields-based localization method, which can be easily implemented in practical engineering applications. A simulation model and a physical experiment are constructed. The simulation and the experiment results provide accurate positioning performance, with the help of verifying the effectiveness of the proposed localization method in underwater target locating.

Active Electro-Location of Objects in the Underwater Environment Based on the Mixed Polarization Multiple Signal Classification Algorithm.

Context plays an important role in the saliency prediction task. In this work, we propose a saliency detection framework that not only extracts visual features but also models two kinds of context including object-object relationships within a single image and scene contextual information. Specifically, we develop a novel saliency detection framework with a Contextual Refinement Module (CRM) which consists of two sub-networks, Object Relation Unit (ORU) and Scene Context Unit (SCU). ORU encodes the object-object relationship based on object relative position and object co-occurrence pattern in an image, by graphical approach, while SCU incorporates the scene contextual information of an image. Object Relation Unit (ORU) and Scene Context Unit (SCU) captures complementary contextual information to give a holistic estimation of salient regions. Extensive experiments show the effectiveness of modelling object relations and scene context in boosting the performance of saliency prediction. In particular, our frame-work outperforms the state-of-the-art models on challenging benchmark datasets.

/pdf/salient-object-detection-by-contextual-refinement-5b9mrqinvn.pdf

Salient Object Detection by Contextual Refinement

An indigenously developed Buried Object Detection SONAR (BODS) for finding targets in the seabed is described in this paper. The important features of BODS are wide bandwidth (2–24 kHz), light weight projector (21 kg) and computer based real time signal processing. The results from experiments at shallow waters to detect seabed boundary layer and targets like metal plate and concrete blocks, are also presented in this paper. It is seen that sediment layers and the targets are distinguishable in the BODS image.

Detection of Buried Objects using active Sonar

This paper focuses on the performance of Multiple Signal Classification (MUSIC) and Estimation of Signal Parameters by Rotational Invariance techniques (ESPRIT) algorithm in real-time, for detection of underwater object by active sonar system of frequency 12 kHz. Experiment is conducted in Acoustic tank facility (ATF) at National Institute of Ocean Technology (NIOT). A table with dimension of 0.5mx0.5m is submerged in the tank at a depth of 3.6m from the water surface. A tow body with acoustic transmitter and hydrophone array is used to transmit and receive the acoustic signal in tank. The DOA estimation gives the exact location of the table in the tank. It is observed that the location of the submerged table computed by MUSIC & ESPRIT is comparable with a bias of about 1 degree.

Experimental observation of direction-of-arrival (DOA) estimation algorithms in a tank environment for sonar application

Context is an important aspect for accurate saliency detection. However, the question of how to formally model image context within saliency detection frameworks is still an open problem. Recent saliency detection models designed using complex Deep Neural Networks to extract robust features, however often fail to select the right contextual features. These methods generally utilize physical attributes of objects for generating final saliency maps, but ignores scene contextual information. In this paper, we overcome such limitation using (i) a proposed novel end-to-end framework with a Contextual Unit (CTU) module that models the scene contextual information to give efficient saliency maps with the help of Convolutional GRU (Conv-GRU). This is the first work reported so far that utilizes Conv-GRU to generate image saliency maps. In addition, (ii) we propose a novel way of using the Conv-GRU that helps to refine saliency maps based on input image context. The proposed model has been evaluated on challenging benchmark saliency datasets, where it outperforms prominent state-of-the-art methods.

Visual Saliency Detection via Convolutional Gated Recurrent Units

National Institute of Ocean Technology (NIOT) Chennai developed a wide band chirp SONAR for buried object detection in shallow waters. It works in the frequency range of 2–24 kHz. It is a towed system. This paper describes indigenously developed SONAR system and the sea trial results, especially the detection of a partially buried ship wreck in shallow water.

Sayanti Bardhan

Papers

Salient Object Detection by Contextual Refinement

Detection of Buried Objects using active Sonar

Experimental observation of direction-of-arrival (DOA) estimation algorithms in a tank environment for sonar application

Visual Saliency Detection via Convolutional Gated Recurrent Units

Effectiveness of Indigenously Developed Chirp Imaging SONAR in Ship Wreck Detection