Vision transformers (ViTs) and convolutional neural networks (CNNs) have demonstrated remarkable performance in classifying complicated hyperspectral images (HSIs). However, these models require a lot of computational power and training data. Recently, Kolmogorov–Arnold Networks (KANs) have been proposed as an effective network to overcome these challenges. In addition to learning new features, KANs may optimize learned features with outstanding accuracy due to their outward similarity to Multi-Layer Perceptrons (MLPs) and internal similarity to splines. Therefore, we introduce a novel gated-adaptable convolutional-based KAN (GC-KAN) for the HSI classification (HSIC). Our method combines a gated approach that selectively focuses on significant features and adaptable convolutional modules that adjust to intricate spectral-spatial changes. Extensive studies on the Houston 2013 and Indian Pines datasets demonstrate the efficacy of GC-KAN, demonstrating its improved performance over conventional techniques. This demonstrates GC-KAN’s potential as an effective tool for more thorough spatial-spectral feature extraction and accurate interpretation for several remote sensing applications.

In this article, a general framework of image annotation is proposed by involving salient object detection (SOD), feature extraction, feature selection, and multi-label classification. For SOD, Augmented-Gradient Vector Flow (A-GVF) is proposed, which fuses benefits of GVF and Minimum Directional Contrast. The article also proposes to control the background information to be included for annotation. This article brings about a comprehensive study of all major feature selection methods for a study on four publicly available datasets. The study concludes with the proposition of using Fisher’s method for reducing the dimension of features. Moreover, this article also proposes a set of features that are found to be strong discriminants by most of the methods. This reduced set for image annotation gives 3--4% better accuracy across all the four datasets. This article also proposes an improved multi-label classification algorithm C-MLFE.

The task of visual question answering (VQA) has gained wide popularity in recent times. Effectively solving the VQA task requires the understanding of both the visual content in the image and the language information associated with the text‐based question. In this study, the authors propose a novel method of encoding the visual information (categorical and spatial object information) of all the objects present in the image into a sequential format, which is called an object sequence. These object sequences can then be suitably processed by a neural network. They experiment with multiple techniques for obtaining a joint embedding from the visual features (in the form of object sequences) and language‐based features obtained from the question. They also provide a detailed analysis on the performance of a neural network architecture using object sequences, on the Oracle task of GuessWhat dataset (a Yes / No VQA task) and benchmark it against the baseline.

Moving object segmentation in maritime domain is a challenging task due to the various real time practical problems, such as waves on the water surface, boat wakes and weather issues (such as bright sun, fog and heavy rain). These problems contribute to generate a highly dynamic background, gradual and sudden illumination changes, camera jitter, shadows and reflections that can provoke false detections. To deal with above issues, in this paper a fast and robust moving object segmentation method on a water surface for maritime surveillance using dynamic background modeling and shadow suppression (OSDBMSS) in the complex wavelet domain is proposed. For dynamic background modeling, we have used frame difference, background registration, background difference and background difference mask in the complex wavelet domain. For shadow detection and removal, we exploit the high frequency subband in the complex wavelet domain. A comparative analysis of the proposed method is presented both qualitatively and quantitatively with other standard methods available in the literature for publicly available datasets of videos in different maritime scenarios, with varying light and weather conditions. Experimental results indicate that the proposed method is performing better in comparison to other standard methods for all the test cases.