Breast cancer has been identified as one of the most common invasive cancers and the second leading cause of cancer death among women. The survival rates have, however, improved dramatically in recent years, thanks to the advances in the screening and treatment process, hugely depending on how early the disease was detected. Along with the physicians, this had also initiated researchers all over the globe to dedicate themselves to extensive research to produce automated diagnosis strategies for breast cancer. Realizing the extraordinary potential of machine learning-based models in the biomedical domain, a large number of diagnosis methods have been proposed in this direction. In our study, we propose a hybrid unique machine learning framework that integrates individual prediction probabilities from 3 machine learning (Logistic Regression, Support Vector Machine, and K Nearest Neighbors) classifiers, then enhances the performance of these 3 classifiers through hybridization, stacking a gradient boosting algorithm over the combination of these classifiers which ultimately results in a 10 Fold Cross Validation Score of 98.4%, Recall of 100% and Precision of 97.3%. Besides, to handle the class imbalance problem we have incorporated SMOTE(Synthetic Minority Oversampling Technique) for minority classes and also Robust Scaling for normalization to deal with outliers in the dataset. In our proposed hybrid solution, we successfully adopted the breast cancer domain in every stage of our framework, starting from data pre-processing, feature extraction and finally classification. Our framework outperformed some recent state of the art studies in the breast cancer domain.
Mixed spectral responses from different ground materials often create confusions in complex remote sensing scenes and restrict classification performance. In this regard, unmixing approaches are being successfully carried out to decompose mixed pixels into a collection of spectral signatures. In this paper, we propose a method to integrate unmixing into a deep feature learning model in order to classify hyperspectral data. We propose to generate superpixels from the abundance estimations of the underlying materials of the image obtained from an unsupervised endmember extraction algorithm called vertex component analysis (VCA). The mean abundances of the superpixels are then used as features for a deep classifier. Our proposed deep model, formulated as a joint convolutional neural network and recurrent neural network, receives significant spectral-spatial information in the data to produce better and powerful features and achieve improved classification performance than several alternative methods.
Convolutional neural networks (CNNs) have demonstrated significant performance in various visual recognition problems in recent years. Recent research has shown that training multilayer neural networks can extensively improve the performance of hyperspectral image (HSI) classification. In this paper, we apply a triplet constraint property on a 3D CNN. This method directly learns a mapping from images to a Euclidean space in which distances directly correspond to a measure of spectral-spatial similarity. Once this embedding has been established, classification can be implemented with such embeddings as feature vectors. Moreover, we also augment the size of the training samples in different band groups. This produces different yet useful estimation of spectral-spatial characteristics of HSI data and contributes considerably in accurate classification. This method is evaluated on a new dataset and compared with several state-of-the-art models, which shows the promising potential of our method.
Image classification is one of the critical tasks in hyperspectral remote sensing. In recent years, significant improvement have been achieved by various classification methods. However, mixed spectral responses from different ground materials still create confusions in complex scenes. In this regard, unmixing approaches are being successfully carried out to decompose mixed pixels into a collection of spectral signatures. Considering the usefulness of these techniques, we propose to utilize the unmixing results as an input to classifiers for better classification accuracy. We propose a novel band group based structure preserving nonnegative matrix factorization (NMF) method to estimate the individual spectral responses from different materials within different ranges of wavelengths. Then we train a convolutional neural network (CNN) with the unmixing results to generate powerful features and eventually classify the data. This method is evaluated on a new dataset and compared with several state-of-the-art models, which shows the promising potential of our method.
This paper proposes a method that uses both spectral and spatial information to segment remote sensing hyperspectral images. After a hyperspectral image is over-segmented into superpixels, a deep Convolutional Neural Network (CNN) is used to perform superpixel-level labelling. To further delineate objects from a hyperspectral scene, this paper attempts to combine the properties of CNN and Conditional Random Field (CRF). A mean-field approximation algorithm for CRF inference is used and formulated with Gaussian pairwise potentials as Recurrent Neural Network. This combined network is then plugged into the CNN which leads to a deep network that has robust characteristics of both CNN and CRF. Preliminary results suggest the usefulness of this framework to a promising extent.
Automated understanding of human facial expression is an active and concerning research topic. It is expected that in near future full-fledged understanding of human facial expression will enable machines to behave more intelligently. In this paper we proposed a system for automatic facial expression recognition. A consistent combination of Self-Organizing Map (SOM), Learning Vector Quantization (LVQ) and Naïve Bayes classifier is developed to recognize facial expression from Cohn Kanade (CK) and Japanese Female Facial Expression (JAFFE) database. Satisfactory experimental results yield the possibility of using this system in real world applications. Proposed methodology shows an accuracy rate of 81.5% for CK dataset and 87.2% accuracy rate for JAFFE dataset.
Road sign recognition is considered to be one of the most fascinating and interesting field of research in intelligent vehicle and machine learning. Road signs are typically placed either by the roadside or above roads. They provide important information in order to make driving safer and easier. This paper proposes an algorithm that recognizes Bangla road sign with a better percentage. The algorithm starts with capture image from real video scene, text detection from images, character segmentation and recognition of characters through shape matrix. The constructed feature vectors for each individual Bangla road sign are learned into a neural network which later classifies new instance of Bangla road sign. The promising preliminary experimental results indicate a positive potential of our algorithm.
In computer vision, semantically accurate segmentation of an object is considered to be a critical problem. The different looking fragments of the same object impose the main challenge of producing a good segmentation. This leads to consider the high-level semantics of an image as well as the low-level visual features which require computationally intensive operations. This demands to optimize the computations as much as possible in order to reduce both computational and communication complexity. This paper proposes a framework which can be used to perform segmentation for a particular object by incorporating optimization in subsequent steps. The algorithm proposes an optimized K-means algorithm for image segmentation followed by balance calculations in multiple instance learning and topological relations with relative positions to identify OOI regions. Finally, a bayesian network is incorporated to contain the learned information about the model of the OOI. The preliminary experimental results suggest a significant drop in the complexity.
Handwritten character recognition is considered to be one of the most fascinating and interesting field of research in image processing and pattern recognition. Due to the various challenges associated with it, intensive research works are currently in progress for constructing algorithms that produce better recognition accuracy. This paper proposes an algorithm that recognizes offline isolated Bangla handwritten characters using spatial relationships between any foreground pixels with the background pixels. The algorithm starts with eliminating unwanted noises from scanned images, performing normalization of size and gradually progress toward constructing feature vector representation for the characters using zoning along with spatial relationships in terms of directional relationships. The constructed feature vectors for each individual Bangla character are learned into a neural network which later classifies new instance of Bangla character. The promising preliminary experimental results indicate a positive potential of our algorithm.
Bangladesh has achieved a tremendous growth in the telecommunication sector recently in spite of various deficiencies regarding socio-economic infrastructure. Following a number of years of huge development in this sector, each and every parts of Bangladesh is now under cellular coverage and the teledensity is expected to reach at maximum satisfactory level soon. This success motivates us towards a more reliable & transparent economic infrastructure through the use of cellular services and intelligent software systems. This paper proposes multilayer network-supported framework which completely eliminates the need and use of paper notes for all kinds of economic transactions. However, the proposed framework is not a threat at all to the present financial institutions and their fundamental mechanisms. Rather, it would be more transparent and free of corruptions as reliable and efficient monitoring of transactions will be possible by respective authorities. This proposed framework will handle all kinds of transactions by electronic means e.g. Push Pull services offering Short Message Service(SMS), road-side booths containing a web interface. Under this distributed system, each and every entity of the entire economic infrastructure will makes it flexible, robust, secure and lawful. The satisfactory experimental results on a small case scenario supports its potential possibilities in real-world implementation.