Remote sensing-induced spectral data is an effective and crucial factor for rice blast monitoring. Recent research focuses on individual spectral data analysis for plant disease forecasting. However, specific spectral data carries particular information for various sites. Hence, only one spectral data would not be identical in recognizing the complex disease pattern in real-time. For better identity and observation, the fusion of various spectral indices while keeping their individuality is necessary and important. Nevertheless, it is challenging to capture the vital cues from the various spectral indices due to failure in explicit feature interactions. To address these issues, a multi granular swin transformer-based model is proposed to mix different spectral indices in a hierarchical way that maintains the robustness of the model. Swin transformer is incorporated to encapsulate the global-local spatial-spectral information from specific spectral images and mixed with another spectral image in the subsequent hierarchy transformer during tokenization. A distinct block (DB) comprised of a multi-head self-attention mechanism and convolution networks is introduced to provide attentive weights of the individual spectral data before mixing. Moreover, the proposed framework captures adaptive spatial-spectral features of rice blast disease from one index and incorporates them with other indices. Extensive experiments are evaluated using the ground truth data over Bardhaman, India and demonstrate the superiority over other state-of-the-art models in terms of performance and model complexity.

Large-scale agricultural production systems require disease monitoring and pest management on a real-time basis. Monitoring disease phenology is one of the possible ways to save agricultural products from huge yield loss incurred due to diseases. Rice is one of the major food crops across the globe. Leaf blast disease in rice affects its productivity all over the world. Monitoring of leaf blast is essential for strategic and tactical disease management decisions. Conventional methods of large-scale disease monitoring are laborious, time taking, and above all, suffer from inaccuracy. Remote sensing parameters are useful for monitoring diseases and crop health on a large scale. Spectral indices derived from remote sensing data provide characteristic features to distinguish areas between healthy and infected crops facilitating strategic application. Assessment of leaf blast incidence based on land surface temperature moderate resolution imaging spectroradiometer (MODIS) and spectral indices normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), normalized difference moisture index (NDMI), soil adjusted vegetation index (SAVI), and moisture stress (Sentinel-2) have been used to predict disease patterns. A deep learning-based model is developed to assess the condition of rice blast disease at field scale. The model provided 90.02% training accuracy and 85.33% validation accuracy. The deep learning model on remote sensing images could assess leaf blast occurrence in real time.

Crowdsourcing is often used as a microtasking platform for solving various complex and monotonic tasks. Notably, in almost every crowdsourcing platform, submission of solution against a requested task is ended with insufficient or no feedback. This might lead to a significant misunderstanding between the requester and crowd workers. In this paper, we address the problem of character recognition with corrective feedback mechanism. This provides a scope to improve the submitted solution and helps in establishing a healthy understanding between the requester and the crowd workers toward quality control. Our results demonstrate that the same mechanism also helps to train new crowd workers in the platform.

This article presents a combinatorial algorithm to find a shortest triangular path (STP) between two points inside a digital object imposed on triangular grid that runs in [Formula: see text] time, where n is the number of pixels on the contour of the object and g is the grid size. Initially, the inner triangular cover which maximally inscribes the object is constructed to ensure that the path lies within the object. An appropriate bounding parallelogram is considered with those two points in diagonally opposite corners and then one of the semi-perimeters of the parallelogram is traversed. Certain combinatorial rules are formulated based on the properties of triangular grid and are applied during the traversal whenever required to shorten the triangular path. A shortest triangular path between any two points may not be unique. Another combinatorial algorithm is presented, which finds the family of shortest triangular path (FSTP) (i.e., the region containing all possible shortest triangular paths) between two given points inside a digital object and runs in [Formula: see text] time. Experimental results are presented to verify the correctness, robustness, and efficacy of the algorithms. STP and FSTP can be useful for shape analysis of digital objects and determining shape signatures. 1

This article presents a combinatorial algorithm to find a shortest triangular path (STP) between two points inside a digital object imposed on triangular grid that runs in O(nglogng)time, where n is the number of pixels on the contour of the object and g is the grid size. Initially, the inner triangular cover which maximally inscribes the object is constructed to ensure that the path lies within the object. An appropriate bounding parallelogram is considered with those two points in diagonally opposite corners and then one of the semi-perimeters of the parallelogram is traversed. Certain combinatorial rules are formulated based on the properties of triangular grid and are applied during the traversal whenever required to shorten the triangular path. A shortest triangular path between any two points may not be unique. Another combinatorial algorithm is presented, which finds the family of shortest triangular path (FSTP) (i.e., the region containing all possible shortest triangular paths) between two given points inside a digital object and runs in O(nglogng) time. Experimental results are presented to verify the correctness, robustness, and efficacy of the algorithms. STP and FSTP can be useful for shape analysis of digital objects and determining shape signatures.

Authorship Attribution is a long-standing problem in Natural Language Processing. Several statistical and computational methods have been used to find a solution to this problem. In this article, we have proposed methods to deal with the authorship attribution problem in Bengali. More specifically, we proposed a supervised framework consisting of lexical and shallow features and investigated the possibility of using topic-modeling-inspired features, to classify documents according to their authors. We have created a corpus from nearly all the literary works of three eminent Bengali authors, consisting of 3,000 disjoint samples. Our models showed better performance than the state-of-the-art, with more than 98% test accuracy for the shallow features and 100% test accuracy for the topic-based features. Further experiments with GloVe vectors [Pennington et al. 2014] showed comparable results, but flexible patterns based on content words and high-frequency words [Schwartz et al. 2013] failed to perform as well as expected.

Granular object segmentation is an important area of image processing, which has several practical applications in agriculture, food industry, geology, and forensics. In this paper, we present a simple algorithm for the analysis of granulometric images that consist of touching or overlapping convex objects such as coffee bean, food grain, nuts, blood cell, or cookies. The algorithm is based on certain underlying digital-geometric features embedded in their binary snapshots. The concept of an outer isothetic cover and the property of geometric convexity are used to extract the joining points (or concavity points) from the ensemble of objects. Next, a combinatorial technique is employed to determine the separator of two overlapping or neighboring objects. This technique is fully automated and it needs only integer-domain computation. The termination time of the algorithm can be traded-off with the quality of segmentation by changing the resolution parameter. Experimental results for a variety of objects chosen from different application domains such as cell image, coffee-bean image and others demonstrate the efficiency and robustness of the proposed method compared to earlier watershed-based algorithms.

A novel algorithm to recognize hand-drawn graphs is proposed. The algorithm uses properties of digital-geometric straightness combined with a new idea of Farey sequence, followed by geometric refinement, in order to speed up the recognition of graph edges. In the next phase, the nodes of the graph - which, being hand-drawn, are very grossly circular - are recognized using the annular regions containing the vertices of their corresponding isothetic covers. Results of the two phases are finally compiled using interval search to output the adjacency list of the graph. The problems of jaggedness, waviness, and similar unforeseen aberrations usually present in a hand-drawn graph are well-tackled by the adopted techniques, as verified by our experimentation on various hand-drawn graphs. Some results have been given in this paper to show the usability and efficiency of the proposed algorithm.

A novel approach towards word segmentation and baseline detection in a handwritten document is proposed. It is based on certain structural properties of isothetic covers tightly enclosing the words in a handwritten document. For an appropriate grid size, the isothetic covers successfully segregates the words so that each cover corresponds to a particular word. By analyzing the horizontal chords of these covers, the corresponding baselines are extracted. The method is fast, robust, and efficient by dint of its traversal strategy along the word boundaries in a combinatorial manner and usage of limited operations strictly in the integer domain. Some results on several Bengali and English handwritings have been given to demonstrate its strength and elegance.

Efficient and successful joining of torn pieces of papers to reconstruct the original documents is an important and challenging issue in many disciplines, especially in forensics and investigation sciences. Automation of the process by means of appropriate techniques can speed up the problem solving substantially. In this paper, we propose a fast, efficient, and useful technique for the reconstruction of hand-torn pages of documents from their images, using contour descriptors for shape-based matching. Chain code of the closed digital arc representing a contour, and its Minkowski sum, have been exploited in our reconstruction work. Experimental results demonstrate the strength and robustness of the method.