The research field of autonomous self-driving vehicles has recently become increasingly popular. In addition, motion-planning technology is essential for autonomous vehicles because it mitigates the prevailing on-road obstacles. Herein, a deep-learning-network-based architecture that was integrated with VGG16 and the gated recurrent unit (GRU) was applied for lane-following on roads. The normalized input image was fed to the three-layer VGG16 output layer as a pattern and the GRU output layer as the last layer. Next, the processed data were fed to the two fully connected layers, with a dropout layer added in between each layer. Afterward, to evaluate the deep-learning-network-based model, the steering angle and speed from the control task were predicted as output parameters. Experiments were conducted using the a dataset from the Udacity simulator and a real dataset. The results show that the proposed framework remarkably predicted steering angles in different directions. Furthermore, the proposed approach achieved higher mean square errors of 0.0230 and 0.0936 and and inference times of 3–4 and 3 ms. We also implemented our proposed framework on the NVIDIA Jetson embedded platform (Jetson Nano 4 GB) and compared it with the GPU’s computational time. The results revealed that the embedded system took 45–46 s to execute a single epoch in order to predict the steering angle. The results show that the proposed framework generates fruitful and accurate motion planning for lane-following in autonomous driving.

Drones are a vital part of our daily lives because of their flexible flying nature and low operation and maintenance costs. Navigation is the most important aspect in the autonomous drone era. With that being said, recent studies have tackled the greatly challenging task of obtaining the most feasible and safest path. In this work, we focus on safe and reliable unmanned aerial vehicle navigation in densely populated city environments to efficiently follow the lane. We present herein a deep learning method through which a drone can autonomously navigate by following a lane in a dynamic urban environment. This drone perceives the environment by using two cameras facing front and down. A convolutional neural network (CNN) is used to capture the high-level feature representation through the attention and residual mechanism of the drone's raw visual inputs. The CNN concurrent pipelines used for three raw visual inputs can output high-level features and follow the concurrent feature fusion into the self-head attention mechanism that encapsulates both dynamic and static information to predict yaw and linear velocity. These predictions are modulated into the final control commands to obtain smooth and continuous inputs that are delivered to the drone. The experimental evaluation results show that the proposed method can learn the drone's navigation strategy for a lane following a guidance system. The high performance of the trained navigation policy is observed in the simulations with different scenarios. In conclusion, our proposed approach outperforms both existing and current state-of-the-art ImageNet models.

Mura defects, which manifest as irregularities in brightness and color on display screens, present persistent challenges for Thin Film Transistor Liquid Crystal Display (TFT-LCD) manufacturers. While traditional methods effectively detect significant abnormalities in vertical-line inspections, identifying minor defects, particularly Level-1, remains formidable. In this study, we leverage artificial intelligence to address this challenge, focusing specifically on detecting vertical-line mura defects. Given the difficulty in discerning Level-1 defects directly from single images, where higher-level abnormalities are more conspicuous, our approach introduces You Only Look Once-Ghost Attention (YOLO-GA). This advanced YOLOv8-based algorithm is meticulously designed to swiftly and accurately identify vertical lines mura in liquid crystal displays (LCD) images, even amidst complex backgrounds and minor irregularities. To enhance the model’s efficacy, we adopt two pivotal strategies. Firstly, we incorporate the Ghost layer as the backbone and neck network to ensure lightweight deployment while improving feature extraction capabilities, particularly in images with intricate backgrounds. Additionally, we integrate the CBAM (Convolutional Block Attention Module) into the network’s architecture, explicitly targeting vertical line mura detection. This augmentation aims to bolster feature extraction and refine detection, especially for defects within liquid crystal displays (LCDs). The dataset utilized in our study is sourced from AUO Corporation and captured using a real single-direction camera. The dataset contains 107,080 images, divided into an 80:20 ratio for training and validation. Each image has a high resolution of 1624 × 1240. Our experimental results show that our approach is highly effective, achieving mAP and F1-Scores of 99.5% and 99.7%, respectively, compared to the baseline model and the comparative attention module. Moreover, the proposed model can significantly reduce the time required to recognize mura defects, fully meeting the manufacturer’s production requirements within a mere 1-second timeframe.

Different tissues and structures in ultrasound produce varying sound wave reflections, resulting in black-and-white grayscale images. Radiologists review thousands of medical images daily and write detailed reports. In parathyroid ultrasound imaging, their reports must go beyond routine observations, providing a comprehensive analysis highlighting the lesion’s type, size, and exact location. This critical information forms the foundation for accurate diagnoses and personalized treatment plans, underscoring the vital role of radiologists in modern medical care. This study presents a novel system for generating automated medical reports from parathyroid ultrasound images focusing on critical diagnostic parameters such as size, type, structure, location, homogeneity, and echogenicity through effective lesion recognition and localization. The proposed system, medical report generation for parathyroid ultrasound (MrGPU-GPT), integrates advanced image processing and natural language generation to streamline clinical workflows and improve diagnostic accuracy. The MrGPU-GPT employs a modified YOLOv7-based architecture, YOLO-BiFPN, to detect and localize lesions within the ultrasound images. Subsequently, key lesion characteristics, including homogeneity and echogenicity, are extracted and utilized for prompt engineering. These prompts are then fed into a fine-tuned Vicuna language model, which generates detailed and precise medical reports. The system provides annotated ultrasound images with bounding boxes and offers comprehensive-textual descriptions in the form of medical reports, significantly reducing the reporting burden on radiologists. Combining image-based diagnostic insights with advanced natural language processing minimizes errors caused by human factors such as inexperience or fatigue, offering robust support for radiologists. The resulting system improves the efficiency and reliability of clinical workflows, delivering accurate diagnostic information while improving patient care outcomes. The experimental results show that, in classifying the lesion, detecting objects, and recognizing position markers, the system achieved a mAP of 93.32%, 91.04%, and 99.65% in the validation of the K-Fold. In medical report generation, compared to MiniGPT, our method, using cosine similarity, ROUGE, and BLEU metrics, along with physician feedback, can generate medical reports with higher similarity scores. (MrGPU-GPT) was deployed on a local computer, developing a medical report in about 3-5 seconds to identify parathyroid images and a medical report.