Breast cancer represents a significant global health challenge, which makes it essential to detect breast cancer early and accurately to improve patient prognosis and reduce mortality rates. However, traditional diagnostic processes relying on manual analysis of medical images are inherently complex and subject to variability between observers, highlighting the urgent need for robust automated breast cancer detection systems. While deep learning has demonstrated potential, many current models struggle with limited accuracy and lack of interpretability. This research introduces the Deep Neural Breast Cancer Detection (DNBCD) model, an explainable AI-based framework that utilizes deep learning methods for classifying breast cancer using histopathological and ultrasound images. The proposed model employs Densenet121 as a foundation, integrating customized Convolutional Neural Network (CNN) layers including GlobalAveragePooling2D, Dense, and Dropout layers along with transfer learning to achieve both high accuracy and interpretability for breast cancer diagnosis. The proposed DNBCD model integrates several preprocessing techniques, including image normalization and resizing, and augmentation techniques to enhance the model’s robustness and address class imbalances using class weight. It employs Grad-CAM (Gradient-weighted Class Activation Mapping) to offer visual justifications for its predictions, increasing trust and transparency among healthcare providers. The model was assessed using two benchmark datasets: Breakhis-400x (B-400x) and Breast Ultrasound Images Dataset (BUSI) containing 1820 and 1578 images, respectively. We systematically divided the datasets into training (70%), testing (20%,) and validation (10%) sets, ensuring efficient model training and evaluation obtaining accuracies of 93.97% for B-400x dataset having benign and malignant classes and 89.87% for BUSI dataset having benign, malignant, and normal classes for breast cancer detection. Experimental results demonstrate that the proposed DNBCD model significantly outperforms existing state-of-the-art approaches with potential uses in clinical environments. We also made all the materials publicly accessible for the research community at: https://github.com/romzanalom/XAI-Based-Deep-Neural-Breast-Cancer-Detection.

This paper presents a real-time computer vision-based Bengali Sign Language (BdSL) recognition system. The system detects the probable hand from the captured image. The system uses Haar-like feature-based cascaded classifiers to detect the hand in each frame. From the detected hand area, the system extracts the hand sign based on Hue and Saturation value corresponding to human skin color. After normalization the system converts the hand sign to binary image. Then the binary images are classified by comparing with pre-trained binary images of hand sign using K-Nearest Neighbors (KNN) Classifier. The system is able to recognize 6 Bengali Vowels and 30 Bengali Consonants. The system is trained using 3600 (36×10×10) training images where each of 10 signers performed 10 signs for each corresponding Bengali alphabet and the system is tested using another 3600 (36×10×10) images of 10 signers. The system is achieved recognition accuracy of 98.17% for Vowels and 94.75% for Consonants.

This paper presents a computer vision based Bengali sign words recognition system using contour analysis. Haar-like feature based cascaded classifier is used to locate the predefined hand posture (Opened Hand and followed by Closed Hand postures) from the captured image, and bounded by a rectangular box that is initialized as region of interest (ROI). The system follows this ROI, crops it and normalizes into predefined size. The system segments skin-like area based on Hue and Saturation value from the normalized image. Then the system employs morphological operations and Gaussian smoothing to remove noises, and then converts it into gray image. The system extracts contours using Canny edge detector and encodes extracted contours into Vector Contours (VC). After scaling VC into predefined size, the system generates feature space based on equalized VC, value of normalized Auto-Correlation Function (ACF) and ACF descriptors for each sign word that will be used for training and/or testing process. The system recognizes sign words based on maximum similarity between tests and predefined training contour templates using Inter-Correlation Function (ICF). The system is trained and tested using 1800 (18×10×10) contour templates separately for 18 Bengali sign words from 10 signers achieving recognition accuracy of 90.11% with computational cost of 26.063 milliseconds per frame.

The usage of Quick Response (QR) Codes has become widely popular in recent years, primarily for immense electronic transactions and industry uses. The structural flexibility of QR Code architecture opens many more possibilities for researchers in the domain of the Industrial Internet of Things (IIoT). However, the limited storage capacity of the traditional QR Codes still fails to stretch the data capacity limits. The researchers of this domain have already introduced different kinds of techniques, including data hiding, multiplexing, data compression, color QR Codes, and so on. However, the research on increasing the data storage capacity of the QR Codes is very limited and still operational. The main objective of this work is to increase the data storage capacity of QR Codes in the IIoT domain. In the first part, we have introduced a 4-State-Pattern-based encoding technique to generate the proposed 4-State QR (4SQR) Code where actual data are encoded into a 4SQR Code image which increases the data storage capacity more than the traditional 2-State QR Code. The proposed 4SQR Code consists of four types of patterns, including Black Square Box (BSB), White Square Box (WSB), Triangle, and Circle, whereas the traditional 2-State QR Codes consist of BSB and WSB. In the second part, the 4SQR Code decoding module has been introduced using the adaptive YOLO V5 algorithm where the proposed 4SQR Code image is decoded into the actual data. The proposed model is tested in a Digital Twin (DT) framework using randomly generated 3000 testing samples for the encoding module that converts into 4SQR Code images successfully and similarly for the decoding module that decodes the 4SQR Code images into the actual data. Experimental results show that this proposed technique offers increased data storage capacity two times than traditional 2-State QR Codes.