Search

This study has a novel approach to capture the attitude of Bottom of the Pyramid (BoP) consumers towards Packaging Influenced Purchase (PIP) during the Covid-19 crisis. Over the years, BoPs consumers have established themselves as an emerging market with ample growth and opportunities. The authors suggested a Multiple-Criteria Decision-Making (MCDM) based framework to assist marketers in targeting both urban and rural BoP consumers regarding PIP. Packaging elements and influence of family, extended family, peers have been included in the framework for gaining in-depth understanding. With a sample size of 100 from West Bengal, this focus group-based study can fulfil the BoP literature’s existing prominent research gap. Results indicate the difference in attitude for urban and rural BoPs towards PIP during this crisis. The fusion of MCDM based approach and relevant machine learning-based technique aims to assist marketers in identifying, formulating, and redefining an action plan.

In this article, an unsupervised domain adaptation strategy has been investigated using a deep Siamese neural network in scene-level land cover classification using remotely sensed images. At the onset, the soft class label and probability scores of each target sample have been obtained using a pretrained model of a deep convolutional neural network. Thereafter, a semiautomatic threshold selection algorithm along with a graph-based approach has been explored to obtain the ``most-confident'' target samples. Furthermore, the deep Siamese network has been incorporated by training the source and ``most-confident'' target samples to generate the classwise cross domain common subspace. To assess the effectiveness of the proposed framework, experiments are carried out using three aerial image datasets. The results are found to be encouraging for the proposed scheme in comparison with the other state-of-art techniques.

Cross-sensor remote sensing data have a significant impact on degrading the performance of traditional land cover classification models. This occurs due to different probability distributions of the collected data from different satellites (having diverse image resolutions and different geographical locations). To resolve this, a cross-sensor domain adaptation (DA) strategy is investigated by considering two source → target scenarios using hyperspectral and aerial image datasets. At the onset, a feature extraction along with a ‘stacking of sample’ (whenever required) strategy is proposed to balance the cross-sensor data in terms of feature dimensions and the available number of samples. Thereafter, a standard deviation-based active learning technique is investigated by exploiting the labeled source images to get the ‘most-informative’ target samples. Finally, the labeled source and ‘most-informative’ target samples are merged to train a classifier which is then used to predict the land cover classes under a multi-sensor framework. Experimental results are found to be promising for the proposed scheme to handle the DA problem under a cross-sensor environment.

A domain adaptation (DA) problem is investigated for land cover classification by utilizing the ensemble decision approach of deep neural networks to address the extra and missing class problem. Two different pretrained models followed by two autoencoders are used to extract the two sets of cross-domain features for samples in the reduced shape. Thereafter, the primary labels and probability scores of target samples are obtained using two different classifiers trained over the source samples with these different features. Moreover, unlike other open set DA techniques, the proposed framework is capable of identifying extra classes separately by exploring the agreement between these two classifiers. Experiments were carried out using three aerial image datasets, and the results are found to be encouraging for the proposed scheme in comparison with other state-of-the-art techniques.

In this letter, a semisupervised neural network-based approach has been proposed for handling the class-imbalance problem in land-cover classification under a hybrid integration of selective undersampling, oversampling, and a bagging-based ensemble of classifiers. Here, a selective undersampling technique is utilized so as to minimize the loss of information from the majority classes; whereas, the minority class sizes are simultaneously increased by exploiting their presence in the unlabeled test samples. Finally, the imbalanced original training set along with the newly found minority samples is used to classify the remaining unlabeled samples from the test set. Experiments conducted on the patterns collected from multispectral (high as well as very high resolution images) and hyperspectral remote sensing satellite images show encouraging performance of the proposed scheme when compared to other state-of-the-art techniques.

In this manuscript, a neuro-fuzzy domain adaptation (DA) technique has been proposed for a multi-level incremental transformation of the source-target features to find an intermediate space with lesser cross-domain distribution difference at each level. In the present investigation, the unsupervised layers of a stacked auto-encoder are used for granular transformation of the weighted samples (or group of samples) at every level. Out of the three, the first two layers of the stack involve unsupervised weighted transformation of source-target samples without using any labelled information from the target domain. After that, a fuzzy membership-based transfer learning scheme has been used to capture the target-distinctive information thereby facilitating a selective transformation between matching source-target sample groups in the third level. Finally, more accurate class-label predictions for the unknown target samples are obtained using the labelled source samples in the transformed (source-target) feature space. To validate the effectiveness of the proposed approach, experimentation has been carried out using samples collected from various multi-spectral satellite images captured over various source and target regions of India. The attained results show superior performance in target class prediction for the proposed DA scheme when compared to other state-of-the-art DA techniques for land-cover classification.

In this letter, a low-cost semisupervised domain adaptation technique has been proposed using a two-level fusion of artificial neural networks. The proposed technique has been aimed to solve the problem of sample selection bias using a one-time collection of a few patterns from the target domain under semisupervised framework. To assess the effectiveness, experiments are conducted on the two source–target data sets acquired over India. The results are found to be encouraging.

In this manuscript, a land-cover classification (LCC) mechanism has been investigated for the practical situations where a remotely sensed aerial image can be annotated by more than one land-cover class. A striking factor for the development of the proposed technique is its ability to operate under situations when there is scarcity of labelled images in the training set, thereby alleviating extensive manual collection of multi-label ground truth information for LCC. The solution using very few labelled training images has been envisioned through a semi-supervised deep learning-based methodology where templates have been generated corresponding to each of the classes present in the training images. Thereafter, each of the test images is assigned to multiple classes based on their similarity with each of the class templates. Experimentation conducted on UCM and AID benchmark multi-label aerial image datasets suggests promising results for the proposed approach.

Advancement in Machine Learning (ML) with improved computing power and large facial image dataset have inspired a lot of research to estimate the age using facial images. These research have also showcased superiority of deeply learned neural networks on large datasets for estimating age. The proposed method utilises the implementation of deep convolutional neural networks (DCNN) using a pre-trained model. It estimates the age using two components called MotherNet (M-Net) and ChildNets (C-Net) which employs a novel dual assurance approach to not only detect but also correct the error in age estimation. Inturn, this strategy improves the overall performance. Experiments have been conducted on two datasets: FG-NET and Chalearn-15. The results show that the proposed approach outperforms the other state-of-the-art methods.

In this manuscript, a deep learning based approach has been investigated for land-cover classification capable of handling situations where there is an imbalance in the class-wise samples in the training set under a novel semi-supervised learning framework. This problem is persistent when the presence of some of the major land-cover classes affects the other recessive classes in a particular region and due to this the generated training set contains a fewer number of samples from the latter group of classes. Here, an adversarial auto-encoder has been used to generate synthetic samples from each of the minority classes so that each of the classes is well represented in the training set. A new training set has then been designed by including the ‘most confident’ artificial samples from the minority classes. This newly formed balanced training set is then shown to be more effective to classify the test samples as compared to the initially available imbalanced training set. The validation of the proposed approach has been performed using patterns collected from two multi-spectral satellite images captured by Ikonos-2 and landsat-8 satellites over different regions of India. The results show significant improvement in test class prediction when compared to that of other state-of-the-art imbalance handling schemes in land-cover classification.

In this article, a domain adaptation (DA) technique using artificial neural networks based classifiers has been proposed using two-level cluster mapping technique by integrating the common data transformation and transfer learning approaches in a single framework. Here, after applying self-organizing feature mapping based clustering technique, a semi-automatic threshold selection mechanism is used to separate out most-confidently paired source–target clusters (i.e. the most similar ones) and alien target clusters (i.e. non-similar ones). Moreover, this strategy makes the proposed technique eligible to apply the transfer learning mechanism. Thereafter, the samples from the most confidently paired target clusters are transformed in terms of the corresponding source clusters using an auto-encoder. Here, the labelled samples are collected from the corresponding source clusters for the paired target clusters; whereas the transfer learning technique is used to select labelled samples from the alien target clusters. To assess the effectiveness of the proposed DA approach, experiments are conducted on the three source–target datasets and the results are compared with other state-of-the-art techniques. Results are also found to be encouraging for the proposed technique.

In this manuscript, a bagging based random undersampling technique using the ensemble of deep convolutional neural networks (DCNNs) has been investigated to deal with the issue of the class-imbalance problem in land-cover classification (LCC). The proposed method is intended to address a scenario in object-level LCC where some land-cover classes outnumber other classes in a specific region, and as a result, the training set contains fewer samples from some (minority) classes compared to other (majority) classes. Here, initially, some bags consisting of balanced classes from the dataset is acquired using random undersampling technique. After that, samples from each bag are used to independently fine-tune a pre-trained convolutional neural network (CNN). Finally, the decisions of all trained finetuned CNN models are combined to assign the class-label for each unlabeled test sample. In order to validate the efficiency of the proposed scheme, experiments have been conducted on the University of California, Aerial Image Dataset, and a new remotely sensed aerial image dataset obtained across the Asian subcontinent region (ASCD). The results obtained over these three different aerial image datasets provide superior performance for the proposed scheme as compared with the other state-of-the-art techniques.

In this manuscript, a feature level ensemble based technique has been proposed using artificial neural networks (ANNs) for an unsupervised domain adaptation (DA) in land-cover classification. Here, ANNs in form of an auto-encoder (AE) and a restricted Boltzman machine (RBM) have been used to learn the sample distribution of both source and target domains together during training. The trained ANNs are then used to find new features suitable for adaptive learning across the two domains having different sample distributions. Finally, the new features obtained from AE and RBM are fused to obtain the final train and test set. Experimentation has been carried out using patterns collected from Landsat-8 multi-spectral satellite images captured over various regions of India. The prediction accuracies of target classes obtained for the proposed scheme show encouraging results when compared with that of other state-of-the-art DA techniques.

Group activity recognition in sports is often challenging due to the complex dynamics and interaction among the players. In this paper, we propose a recurrent neural network to classify puck possession events in ice hockey. Our method extracts features from the whole frame and appearances of the players using a pre-trained convolutional neural network. In this way, our model captures the context information, individual attributes and interaction among the players. Our model requires only the player positions on the image and does not need any explicit annotations for the individual actions or player trajectories, greatly simplifying the input of the system. We evaluate our model on a new Ice Hockey Dataset. Experimental results show that our model produces competitive results on this challenging dataset with much simpler inputs compared with the previous work.

In this article, a land-cover classification technique is proposed using semi-supervised Multilayer Perceptron (MLP) where only a few labeled samples are available to train the classifier. Here, the MLP initially trained using a few labeled samples is used to predict the class labels of the unlabeled samples to obtain soft class labels. The soft class labels of each unlabelled samples are boosted with respect to that of their k nearest neighbors. This is done to exploit the spatial relationship among the patterns. Finally re-training for MLP is carried out iteratively by using the available labeled samples along with the selected samples from the most confidently predicted ones. The retraining continues until the prediction values are significantly stable across iterations. Here a differential weighted mean of k neighborhood is used for boosting of soft class label and a semi-automatic threshold selection algorithm is investigated to identify the most confidently predicted samples. To assess the effectiveness of the proposed methodology, experiments on land-cover classification are conducted on patterns from Landsat8 satellite images for two different geographical regions. The obtained results outperform the corresponding supervised strategy under the scenario of labeled data scarcity.

In this article, a novel approach using ensemble of semi-supervised classifiers is proposed for change detection in remotely sensed images. Unlike the other traditional methodologies for detection of changes in land-cover, the present work uses a multiple classifier system in semi-supervised (leaning) framework instead of using a single weak classifier. Iterative learning of base classifiers is continued using the selected unlabeled patterns along with a few labeled patterns. Ensemble agreement is utilized for choosing the unlabeled patterns for the next training step. Finally, each of the unlabeled patterns is assigned to a specific class by fusing the outcome of base classifiers using some combination rule. For the present investigation, multilayer perceptron (MLP), elliptical basis function neural network (EBFNN) and fuzzy k-nearest neighbor (k-nn) techniques are used as base classifiers. Experiments are carried out on multi-temporal and multi-spectral images and the results are compared with the change detection techniques using MLP, EBFNN, fuzzy k-nn, unsupervised modified self-organizing feature map and semi-supervised MLP. Results show that the proposed work has an edge over the other state-of-the-art techniques for change detection.

In the present article, semi-supervised learning is integrated with an unsupervised context-sensitive change detection technique based on modified self-organizing feature map (MSOFM) network. In the proposed methodology, training of the MSOFM network is initially performed using only a few labeled patterns. Thereafter, the membership values, in both the classes, for each unlabeled pattern are determined using the concept of fuzzy set theory. The soft class label for each of the unlabeled patterns is then estimated using the membership values of its K nearest neighbors. Here, training of the network using the unlabeled patterns along with a few labeled patterns is carried out iteratively. A heuristic method has been suggested to select some patterns from the unlabeled ones for training. To check the effectiveness of the proposed methodology, experiments are conducted on three multi-temporal and multi-spectral data sets. Performance of the proposed work is compared with that of two unsupervised techniques, a supervised technique and two semi-supervised techniques. Results are also statistically validated using paired t-test. The proposed method produced promising results.

In this letter, a change detection technique using a multiple classifier system is proposed. Here, different architectures of multilayer perceptron (MLP) are used as base classifiers. An ensemble of different MLPs is utilized to increase the robustness of the system. This also avoids the problem of choosing an optimum architecture for MLP. First, the support values for each of the unlabeled patterns are estimated using different MLPs (trained with the labeled patterns). Then, each of the unlabeled patterns is assigned to a specific class by fusing the outcome of the base classifiers using different combination rules. Experiments are carried out on multitemporal and multispectral images. Results show that the proposed ensemble technique has an edge over individual base classifiers for change detection in remotely sensed images.

Traffic signs should be accurately identified in order to prevent vital road accidents and secure lives. The objective of this paper is to detect the warning traffic signs and recognize the message it is designed to convey. This system is based on extracting the warning sign from the traffic scene by Windowed Hough Transform. Next Histogram Oriented Gradient (HOG) is used to collect the feature of the extracted part of triangular object and finally SVM classifier is applied to train the HOG features. To summarize, the system first detects the warning traffic sign in the first place, specifies whether the detected sign is a warning sign, and then determines the meaning of the symbol inside it. The SVM classifier was trained with 200 images which were collected in different light conditions. To check the robustness of this system, it was tested against 327 images which contain 292 warning traffic sign and 35 other types of traffic signs. It was found that the accuracy of recognition was approximately 94% which indicates clearly the high robustness targeted by this system.

Domain Adaptation (DA) for remotely sensed images using ensemble of Multilayer Perceptrons (MLP) is presented in this artide. Using the labelled information from a source domain, the proposed method utilises the disagreement among the MLPs to figure out the ‘most informative’ patterns from the target domain using transfer learning. These selected patterns are then labelled by human expert and are used for training the ensemble. Finally the land-cover classes for a target region are predicted by applying a non-trainable majority voting rule among different MLPs in ensemble. Experiments have been conducted on multispectral images for two different regions in India obtained from Landsat-8 satellite and the proposed DA method outperforms the corresponding random sampling approach.

Online Social Networks (OSN), such as Facebook, Twitter, Youtube and so on, are important sources of online content today. These platforms are used by millions of people world-wide, to share information and express their sentiment and opinion on various social issues. Sentiment analysis of online content – automatically inferring whether a particular textual content reflects a positive (e.g., happy) or negative (e.g., sad) sentiment of the person who posted the content – is an important research problem today, and has several potential applications such as analysing public opinion on various products or social issues. In this paper, we propose a simple but effective methodology of inferring the sentiment of textual content posted in online social media. Our approach is based on first identifying the positive / negative polarity of terms, i.e., whether a certain term (e.g., a word) is normally used in a positive or negative context, and then to infer the sentiment of a given text based on the polarity of the terms present in the text. A key challenge in this approach is that in online social media, different users use different words while expressing similar opinion. To address this, we use the well-known lexical database WordNet to identify groups of words which are synonymous to each other. We apply our proposed methodology on a large publicly available dataset containing content from six different online social media, which has been labeled as positive / negative by human annotators, and find that our methodology achieves better performance than several approaches developed earlier.

The Twitter microblogging site is one of the most popular websites in the Web today, where millions of users post real-time messages (tweets) on different topics of their interest. The content that becomes popular in Twitter (i.e., discussed by a large number of users) on a certain day can be used for a variety of purposes, including recommendation of popular content and marketing and advertisement campaigns. In this scenario, it would be of great interest to be able to predict what content will become popular topics of discussion in Twitter in the recent future. This problem is very challenging due to the inherent dynamicity in the Twitter system, where topics can become hugely popular within short intervals of time. The Twitter site periodically declares a set of trending topics, which are the keywords (e.g., hashtags) that are at the center of discussion in the Twitter network at a given point of time. However, the exact algorithm that Twitter uses to identify the trending topics at a certain time is not known publicly. In this paper, we aim to predict the keywords (hashtags) that are likely to become trending in Twitter in the recent future. We model this prediction task as a machine learning classification problem, and analyze millions of tweets from the Twitter stream to identify features for distinguishing between trending hashtags and non-trending ones. We train classifiers on features measured over one day, and use the classifiers to distinguish between trending and non-trending hashtags on the next day. The classifiers achieve very high precision and reasonably high recall in identifying the hashtags that are likely to become trending.

In the present article, we have provided a performance analysis of different classifier-based object tracking techniques. Here, object tracking has been considered as a binary classification problem. Different classifiers used in the present work are k-nearest neighbor (k-NN), fuzzy k-nearest neighbor (fuzzy k-NN), multilayer perceptron (MLP) neural network, and radial basis function (RBF) neural network. The object scale changes are controlled by considering an adaptive tracking technique. The present work is tested with different video sequences. However, for page constraints, we have provided results on two benchmark video sequences only. The performance of different object tracking techniques were evaluated by two evaluation measures: overlapped area and centroid distance.

In this article, a one-dimensional self-organizing feature map (SOFM) neural network integrated with semi-supervised learning is used to predict the class label of gene expression data under the scarcity of the labeled patterns. Iterative learning of the semi-supervised SOFM network is carried out using a few labeled patterns along with some selected unlabeled patterns. The unlabeled patterns, for which the maximum target value is greater than a threshold, are selected as the confident ones. Results are found to be encouraging.

In the proposed work, a change detection technique is developed using a combination of multilayer perceptrons (MLPs). At the onset, the different MLPs are trained with the labeled patterns. Then, the support values (or, the output values) for the unlabeled patterns are obtained from these trained MLPs. At last, decision regarding the class assignment for the unlabeled patterns has been made by fusing the outcome (i.e., support values) obtained from different trained MLPs. In the present experiment, ‘mean rule’ and ‘majority voting’ are used as combination rules. Experiments are carried out on multi-temporal and multi-spectral remotely sensed images. Results for the proposed methodology are found to be encouraging.

In the present work, a change detection technique in remotely sensed images (under the scarcity of labeled patterns) is proposed where an ensemble of semi-supervised classifiers is used, instead of using a single (weak) classifier. Iterative learning of multiple classifier system is carried out using the selected unlabeled patterns along with a few labeled patterns. Selection of unlabeled patterns for the next training step is done using ensemble agreement. Finally, the unlabeled patterns are assigned to a class by fusing the outcome of base classifiers using a combiner. For the present investigation, multilayer perceptron (MLP), elliptical basis function neural network (EBFNN) and fuzzy k-nearest neighbor (KNN) techniques are used as base classifiers. Experiments are carried out on multi-temporal and multi-spectral images and the results for the proposed methodology are found to be encouraging.

In real life change detection for remotely sensed images suffers due to the problem of inadequate labeled patterns. When a few labeled patterns can be collected by experts, semi-supervised (learning) clustering can be opted for change detection instead of the unsupervised approach to make full utilization of both labeled and unlabeled patterns. In the present work, a study has been carried out by applying some of the semi-supervised clustering techniques for changed detection. A comparative analysis between K-Means, COP-KMeans, Seeded-KMeans and Constrained-KMeans algorithms is being performed based on the results obtained using two multi-temporal remotely sensed images. It can be concluded from the experiments that the Constrained-KMeans is well suited for changed detection of remotely sensed images under semi-supervised framework.

In this article, we propose a change detection technique using semi-supervised Hopfield-Type Neural Network (HTNN). The purpose of the work is to show the usefulness of semi-supervision over existing unsupervised/fully supervised methods when we have only a few labeled samples. Here, training of HTNN is performed iteratively using a few labeled patterns along with a number of unlabeled patterns. A method has been suggested to propagate the label information using a kind of K-nearest neighbor approach. To check the effectiveness of the proposed method, experiments are carried out on multi-temporal remotely sensed images. Results are compared with other state of the art techniques and found to be significantly better.

Problem of change detection of remotely sensed images using insufficient labeled patterns is the main topic of present work. Here, semisupervised learning is integrated with an unsupervised context-sensitive change detection technique based on modified self-organizing feature map (MSOFM) network. In this method, training of theMSOFMis performed iteratively using unlabeled patterns along with a few labeled patterns. A method has been suggested to select unlabeled patterns for training. To check the effectiveness of the proposed methodology, experiments are carried out on two multitemporal remotely sensed images. Results are found to be encouraging.

Scarcity of sufficient ground truth information is the primary bottleneck for adopting any supervised methodology in change detection domain and hence, unsupervised approaches are mostly used for this task. But, with a few labelled patterns in hand, semi-supervised methods can be chosen instead of unsupervised ones to utilise both the labelled and unlabelled patterns completely. Work on semi-supervised learning both in the areas of clustering and classification is now being explored. In this article, a detailed study has been made by applying some of the semi-supervised clustering techniques for change detection. In present investigation, five semi-supervised clustering techniques, namely COP-KMeans, seeded-KMeans, constrained-KMeans, semi-supervised-HMRF-KMeans and semi-supervised-kernel-KMeans algorithms are used. A comparative analysis has been made among these algorithms and standard K-Means algorithm, using two multi-temporal remotely sensed images and are also statistically validated using paired t-test. Experimental results conclude that constrained-KMeans for both the datasets is more applicable for change detection than COP-KMeans and seeded-KMeans. Semi-supervised-HMRF-KMeans and semi-supervised-kernel-KMeans algorithms are found not to be robust for all the datasets because these algorithms outperform constrained-KMeans in case of only one dataset.