Emotion is a scorching topic in the recent years due to the critical unseen stress incurred during the pandemic and post-pandemic. This is worsening with the recent economy’s inflation and increase of living cost, many employees are seriously affected and drawn forth many families saddened cases and tremendous drop of working performance. The increasing stress brings a lot of harm not only to the individual but to the company’s and country’s growth. To recognize emotion through a single model is less accurate, however, recruiting multiple-models may lead to latency in data processing and possibly misleading results if the input models data are not properly filtered and segmented. This paper will review, analyze and theoretically compare 15 facial expression methods and 17 voice methods of emotion recognition research works. It will outline the pros and cons of each method and discuss the accuracy of some of the standalone and hybrid emotion recognition methods. Some of the methods (such as CNN, KNN and SVM) can span over multiple-models, but reveal different level of strengths. This is very important to discover, so that one may replace or enhance the weaker level if applying the same method across the multiple-models. This paper will also illustrate different levels of popularity of the methods in each model for visual comparison in ease. Hopefully, it can cater the new researchers a quick identification on the most suitable method for recognizing the emotion through facial expression and/or voice.

The presence issue of inaccurate plant disease detection persists under real field conditions and most deep learning (DL) techniques still struggle to achieve real-time performance. Hence, challenges in choosing a suitable deep-learning technique to tackle the problem should be addressed. Plant diseases have a detrimental effect on agricultural yield, hence early detection is crucial to prevent food insecurity. To identify and categorise the indications of plant diseases, numerous developed or modified DL architectures are utilised. This paper aims to observe the performance of the YOLOv8 model, which has better performance than its predecessors, on a small-scale plant disease dataset. This paper also aims to improve the accuracy and efficiency of plant disease detection and classification methods by proposing an optimised and lightweight YOLOv8 architecture model. It trains the YOLOv8 model on a public dataset and optimises the YOLOv8 algorithm with the integration of the GhostNet module into the backbone architecture to cut down the number of parameters for a faster computational algorithm. In addition, the architecture incorporates a Coordinate Attention (CA) mechanism module, which further enhances the accuracy of the proposed algorithm. Our results demonstrate that the combination of YOLOv8s with CA mechanism and transfer learning obtained the best result, yielding score of 72.2% which surpassed the studies that utilised the same dataset. Without transfer learning, our best result is demonstrated by YOLOv8s with GhostNet and CA mechanism yielding a score of 69.3%.

Many blind individuals have difficulties in recognizing people’s facial expression which may impact their social interaction. With the recognition, the blind individuals can accurately interpret and respond to the emotions. There is a lack in the existing application with the combination of face and facial expressions recognition. The blind individuals have to rely on multiple applications to accomplish the same task, making it difficult and time-consuming for them to use. The paper aims to recognize faces and facial expressions for blind individuals and provides feedback in real-time. Three face detection algorithms of Haar Cascade Classifier, Dlib, and RetinaFace are compared. Dlib is chosen to process with Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM). It loads the pre-trained model, computes the HOG features, slide the window scanning at different scales, classify the windows using the SVM classifier, generate bounding boxes, and applying non-maximum suppression. ResNet50 architecture is employed to recognize face and Convolutional Neural Networks (CNN) is applied to recognize facial expression. The training accuracy is 70% and validation accuracy is 60%.

The tremendous development in Natural Language Processing (NLP) has enabled the detection of bilingual and multilingual textual similarity. One of the main challenges of the Textual Similarity Detection (TSD) system lies in learning effective text representation. The research focuses on identifying similar texts between Indonesian and English across a broad range of semantic similarity spectrums. The primary challenge is generating English and Indonesian dense vector representation, a.k.a. embeddings that share a single vector space. Through trial and error, the research proposes using the Universal Sentence Encoder (USE) model to construct bilingual embeddings and FAISS to index the bilingual dataset. The comparison between query vectors and index vectors is done using two approaches: the heuristic comparison with Euclidian distance and a clustering algorithm, Approximate Nearest Neighbors (ANN). The system is tested with four different semantic granularities, two text granularities, and evaluation metrics with a cutoff value of k={2,10}. Four semantic granularities used are highly similar or near duplicate, Semantic Entailment (SE), Topically Related (TR), and Out of Topic (OOT), while the text granularities take on the sentence and paragraph levels. The experimental results demonstrate that the proposed system successfully ranks similar texts in different languages within the top ten. It has been proven by the highest F1@2 score of 0.96 for the near duplicate category on the sentence level. Unlike the near-duplicate category, the highest F1 scores of 0.77 and 0.89 are shown by the SE and TR categories, respectively. The experiment results also show a high correlation between text and semantic granularity.