The intrusion of foreign objects on airport runways during aircraft takeoff and landing poses a significant safety threat to air transportation. Small-scale Foreign Object Debris (FOD) cannot be ruled out on time by traditional manual inspection, and there is also a potential risk of secondary foreign body intrusion. A deep-learning-based intelligent detection method is proposed to solve the problem of low accuracy and low efficiency of small-scale FOD detection. Firstly, a dual light camera system is utilized for the collection of FOD data. It generates a dual light FOD dataset containing both infrared and visible light images. Subsequently, a multi-attention mechanism and a bidirectional feature pyramid are integrated into the baseline network YOLOv5. This integration prioritizes the extraction of foreign object features and boosts the network’s ability to distinguish FOD from complex backgrounds. Additionally, it enhances the fusion of higher-level features to improve the representation of multi-scale objects. To ensure fast and accurate localization and recognition of targets, the Complete-IoU (CIoU) loss function is used to optimize the bounding boxes’ positions. The experimental results indicate that the proposed model achieves a detection speed of 36.3 frame/s, satisfying real-time detection requirements. The model also attains an average accuracy of 91.1%, which is 7.4% higher than the baseline network. Consequently, this paper verifies the effectiveness and practical utility of our algorithm for the detection of small-scale FOD targets.

This study aims to enhance pavement crack detection methods by integrating unmanned aerial vehicles (UAVs) with deep learning techniques. Current methods encounter challenges such as low accuracy, limited efficiency, and constrained application scenarios. We introduce an innovative approach that employs a UAV equipped with a binocular camera for identifying pavement surface cracks. This method is augmented by a binocular ranging algorithm combined with edge detection and skeleton extraction algorithms, enabling the quantification of crack widths without necessitating a preset shooting distance—a notable limitation in existing UAV crack detection applications. We developed an optimized model to enhance detection accuracy, incorporating the YOLOv5s network with an Efficient Channel Attention (ECA) mechanism. This model features a decoupled head structure, replacing the original coupled head structure to optimize detection performance, and utilizes a Generalized Intersection over Union (GIoU) loss function for refined bounding box predictions. Post identification, images within the bounding boxes are segmented by the Unet++ network to accurately quantify cracks. The efficacy of the proposed method was validated on roads in complex environments, achieving a mean Average Precision (mAP) of 86.32% for crack identification and localization with the improved model. This represents a 5.30% increase in the mAP and a 6.25% increase in recall compared to the baseline network. Quantitative results indicate that the measurement error margin for crack widths was 10%, fulfilling the practical requirements for pavement crack quantification.

Demosaicing technology stands as one of the most critical components in polarization filter arrays (PFAs) image processing, directly determining the quality of reconstructed information from the division of focal plane (DoFP) polarization imaging system. To address the challenges of large instantaneous field-of-view (IFoV) errors and poor polarization reconstruction accuracy in DoFP polarization imaging systems, we introduce two novel prior constraints, to our knowledge, related to the system’s extinction ratio and non-uniformity based on PFA superpixel methodology. Through systematic calibration of the imaging system to obtain these prior parameters, we propose a superpixel Stokes vector frequency reduction approach and validate its effectiveness through comparative analysis with conventional polarization difference models. The method employs inverse distance weighted (IDW) interpolation for Stokes vector reconstruction at the superpixel level, enabling retrieval of all four angular measurements per pixel to complete the demosaicing process. Subjective visual evaluations and objective quantitative metrics demonstrate that our algorithm achieves comparable performance to state-of-the-art methods while offering closer approximation to ground truth polarization parameters, reduced computational complexity, and enhanced suitability for embedded engineering implementation.