Abstract This study focuses on real-time hand gesture recognition in the Turkish sign language detection system. YOLOv4-CSP based on convolutional neural network (CNN), a state-of-the-art object detection algorithm, is used to provide real-time and high-performance detection. The YOLOv4-CSP algorithm is created by adding CSPNet to the neck of the original YOLOv4 to improve network performance. A new object detection model has been proposed by optimizing the YOLOv4-CSP algorithm in order to provide more efficient detection in Turkish sign language. The model uses CSPNet throughout the network to increase the learning ability of the network. However, Proposed YOLOv4-CSP has a learning model with Mish activation function, complete intersection of union (CIoU) loss function and transformer block added. The Proposed YOLOv4-CSP algorithm has faster learning with transfer learning than previous versions. This allows the proposed YOLOv4-CSP algorithm to perform a faster restriction and recognition of static hand signals simultaneously. To evaluate the speed and detection performance of the proposed YOLOv4-CSP model, it is compared with previous YOLO series, which offers real-time detection, as well. YOLOv3, YOLOv3-SPP, YOLOv4-CSP and proposed YOLOv4-CSP models are trained with a labeled dataset consisting of numbers in Turkish Sign language, and their performances on the hand signals recognitions are compared. With the proposed method, 98.95% precision, 98.15% recall, 98.55 F1 score and 99.49% mAP results are obtained in 9.8 ms. The proposed method for detecting numbers in Turkish sign language outperforms other algorithms with both real-time performance and accurate hand sign prediction, regardless of background.

Abstract Medical datasets often have a skewed class distribution and a lack of high-quality annotated images. However, deep learning methods require a large amount of labeled data for classification. In this study, we present a few-shot learning approach for the classification of ultrasound breast cancer images using meta-learning methods. We used prototypical networks and model agnostic meta-learning (MAML) algorithms as meta-learning methods. The breast ultrasound images (BUSI) dataset, which has three classes and is difficult to use in meta-learning, was used for meta-testing in a cross-domain approach along with other datasets for meta-training. Our proposed approach yielded an accuracy range of 0.882–0.889, achieved by implementing the ResNet50 backbone with ProtoNet in a 10-shot setting. These results represent a significant improvement ranging from 6.27 to 7.10% over the baseline accuracy of 0.831. The results showed that ProtoNet outperformed the MAML method for all k-shot settings. In addition, the use of ResNet models as the backbone network for feature extraction was found to be more successful than the use of a four-layer convolutional model. Our proposed method is the first attempt to apply meta-learning for few-shot classification in the BUSI dataset while providing higher accuracy compared to deep learning methods for medical images with small-scale datasets and few classes. The methodology used in this study can be adapted to other datasets with similar problems.