Oppositional Tuna Swarm Optimization with Convolutional Neural Network (OTSO-CNN) classifier for American Sign Language recognition

Background For those with hearing loss, sign language (SL) is an essential communication tool. The most common and basic type of SL is American SL (ASL). ASL plays a vital role in various communicative contexts, including spelling names, correcting letters, and reading books. Accurate real-time (RT) SL recognition is essential. Image recognition (IR) challenges that are pertinent to this objective have demonstrated the enormous promise of deep learning (DL) approaches, especially convolutional neural networks (CNNs). But conventional CNN frameworks struggle with understanding inter-image connections due to their nonlinear feature transformations. Methods This study proposes a DL-based system using the ASL Alphabet (ASLA) dataset to recognize ASL signs. The system employs the You Only Look Once v7 (YOLOv7) architecture to detect and localize the bounding boxes of hand signs in images, effectively reducing background interference and enhancing sign focus. To further optimize classification, the model integrates Oppositional Tuna Swarm Optimization (OTSO) with CNNs. OTSO optimizes CNN parameters using two foraging behavior models spiral and parabolic based on tuna behavior. This approach ensures efficient learning and adaptability of the CNN classifier tailored to the ASLA dataset. Results The proposed OTSO-CNN model demonstrates robust performance in real-time ASL recognition. It continuously received high scores on every metric used for evaluation, including precision (P), recall (R), F1-score, and accuracy (ACC). This confirms the system’s effectiveness and potential usability by individuals with hearing impairments for reliable and accurate ASL communication.