Various technological devices have been developed to meet the ever-increasing demand of today’s cities. Cities, as we can see today, have become highly technology-oriented. However, technological advancement comes at a cost which in this case is the environment of our planet. Therefore, it is necessary to design green cities, which impose as less harm as it possibly can to the environment. One of the most important characteristics of these new cities is the way of managing their waste. Traditional waste management, which employs various sizes and shapes of trash cans at multiple places that are collected by hand after each use is space-consuming, manual, inefficient and often leads to environmental pollution. Therefore, we require a novel approach that is free from these constraints. This paper proposes a garbage management system integrating the fundamental ideas of smart waste management and ‘Internet of Things’ (IoT). The proposed system employs smart cyclic containers that rotate one by one after being filled with rubbish like ‘Merry Go Round’. This approach of waste management solves the problem of space-constraint in an innovative way. A central server monitors the whole system, which further dispatches an autonomous car to collect the waste when necessary. This IoT integrated waste management system is named ‘Internet of Garbage Bins’ (IoGB).

Sign language is undoubtedly a common way of communication among deaf and non-verbal people. But it is not common among hearing people to use sign language to express feelings or share information in everyday life. Therefore, a significant communication gap exists between deaf and hearing individuals, despite both groups experiencing similar emotions and sentiments. In this paper, we developed a convolutional neural network–squeeze excitation network to predict the sign language signs and developed a smartphone application to provide access to the ML model to use it. The SE block provides attention to the channel of the image, thus improving the performance of the model. On the other hand, the smartphone application brings the ML model close to people so that everyone can benefit from it. In addition, we used the Shapley additive explanation to interpret the black box nature of the ML model and understand the models working from within. Using our ML model, we achieved an accuracy of 99.86% on the KU-BdSL dataset. The SHAP analysis shows that the model primarily relies on hand-related visual cues to predict sign language signs, aligning with human communication patterns.

Sign language is a form of communication medium for speech and hearing disabled people. It has various forms with different troublesome patterns, which are difficult for the general mass to comprehend. Bengali sign language (BdSL) is one of the difficult sign languages due to its immense number of alphabet, words, and expression techniques. Machine translation can ease the difficulty for disabled people to communicate with generals. From the machine learning (ML) domain, computer vision can be the solution for them, and every ML solution requires a optimized model and a proper dataset. Therefore, in this research work, we have created a BdSL dataset and named `KU-BdSL', which consists of 30 classes describing 38 consonants (‘banjonborno’) of the Bengali alphabet. The dataset includes 1500 images of hand signs in total, each representing Bengali consonant(s). Thirty-nine participants (30 males and 9 females) of different ages (21–38 years) participated in the creation of this dataset. We adopted smartphones to capture the images due to the availability of their high-definition cameras. We believe that this dataset can be beneficial to the deaf and dumb (D&D) community. Identification of Bengali consonants of BdSL from images or videos is feasible using the dataset. It can also be employed for a human-machine interface for disabled people. In the future, we will work on the vowels and word level of BdSL.

Sign language is the most prevalent form of communication among people with speech and hearing disabilities. The most widely used types of sign language involve the creation of static or dynamic gestures using hand(s). Among many sign languages, Bengali Sign Language (BdSL) is one of the most complicated sign languages to learn and comprehend because of its enormous alphabet, vocabulary, and variation in expression techniques. Existing solutions include learning BdSL or hiring an interpreter. Besides, BdSL interpreter support is hard to come by and expensive (if not voluntary). Disabled people might find it more comfortable to converse with generals implementing machine translation of sign language. Deep learning that mimics the human brain, a subset of the machine learning domain, seems to be a viable solution. For the deaf and dumb (D&D) community, computer vision, in particular, may hold the key to finding a solution. Therefore, we have created a novel model, KUNet ("Khulna University Network" a CNN based model), a classification framework optimized by the genetic algorithm (GA), has been proposed to classify BdSL. This model and the dataset contribute to creating a BdSL machine translator. GA-optimized KUNet acquired an accuracy of 99.11% on KU-BdSL. After training the model on KU-BdSL, we demonstrated a comparison of the model with state-of-the-art studies and interpreted the black-box nature of the model using explainable AI (XAI). Additionally, we have found that our model outperformed several well-known models trained on the KU-BdSL dataset. This study will benefit the D&D community by allowing them to communicate effortlessly and minimizing their hardship.

EEG signals can assess human emotions that perform significantly in promoting robust brain-computer interface systems. This study has been conducted utilizing a public emotional EEG dataset called SEED. Before performing classification from the EEG signals, features have been extracted to acquire information. DWT has been introduced to disintegrate the preprocessed EEG signals into five separate frequency sub-bands (alpha, beta, gamma, delta, and theta), and the statistical features of these DWT coefficients have been computed in the frequency domain. We applied three different wavelet functions to extract multiple features from the preprocessed signals and considered three statistical ones to distinguish relevant information about EEG signals. The wavelet functions are “coif5”, “db4”, and “db8”, and the statistical features are entropy, power, and standard deviation. In this experiment, the efficiency of emotion identification has been investigated for a set of 62 EEG channels. A non-linear SVM classifier has been trained using the features to categorize the signals into three mental states: negative, neutral and positive by applying 5-fold cross-validation. In our model, we achieved a training accuracy of 70%, 83.75% and 82%on 62 channels for positive, neutral and negative, respectively.

The KU-BdSL refers to a Bengali sign language dataset, which includes three variants of the data. The variants are - (i) Uni-scale Sign Language Dataset (USLD), (ii) Multi-scale Sign Language Dataset (MSLD), and (iii) Annotated Multi-scale Sign Language Dataset (AMSLD). The dataset consists of images representing single-hand gestures for BdSL alphabets. Several smartphones are taken into account to capture images from 33 participants (25 males and 8 females). Each version includes 30 classes that resemble the 39 consonants ('shoroborno') of Bengali alphabets. There is a total of 1,500 images in jpg format in each variant. The images are captured on flat surfaces at different times of the day to vary the brightness and contrast. Class names are Unicode values corresponding to the Bengali alphabets for USLD and MSLD. Folder Names: 2433 -> ‘Chandra Bindu’ 2434 -> ‘Anusshar’ 2435 -> ‘Bisharga’ 2453 -> ‘Ka’ 2454 -> ‘Kha’ 2455 -> ‘Ga’ 2456 -> ‘Gha’ 2457 -> ‘Uo’ 2458 -> ‘Ca’ 2459 -> ‘Cha’ 2460-2479 -> ‘Borgio Ja/Anta Ja’ 2461 -> ‘Jha’ 2462 -> ‘Yo’ 2463 -> ‘Ta’ 2464 -> ‘Tha’ 2465 -> ‘Da’ 2466 -> ‘Dha’ 2467-2472 -> ‘Murdha Na/Donto Na’ 2468-2510 -> ‘ta/Khanda ta’ 2469 -> ‘tha’ 2470 -> ‘da’ 2471 -> ‘dha’ 2474 -> ‘pa’ 2475 -> ‘fa’ 2476-2477 -> ‘Ba/Bha’ 2478 -> ‘Ma’ 2480-2524-2525 -> ‘Ba-y Ra/Da-y Ra/Dha-y Ra’ 2482 -> ‘La’ 2486-2488-2487 -> ‘Talobbo sha/Danta sa/Murdha Sha’ 2489 -> ‘Ha’ USLD: USLD has a unique size for all the images that is 512*512 pixels. The intended hand position is placed in the middle of the majority of cases in this dataset. MSLD: The raw images are stored in MSLD so that researchers can make changes to the dataset. The use of various smartphones yields us a wide variety of image sizes. AMSLD: AMSLD has multi-scale annotated data, which is suitable for tasks like localization and classification. From many annotation formats, the YOLO DarkNet annotation has been selected. Each image has an annotation text file containing five numbers separated by white space. The initial number is an integer, and the rest are floating numbers. The first number of the file indicates the class ID corresponding to the label of that image. Class IDs are mapped in a separate text file named 'obj.names'. The second and third values are the beginning normalized coordinates, while the fourth and fifth define the bounding box's normalized width and height. All the data from this dataset is free to download, modify, and use.