Abstract:Foreign Object Debris (FOD) is a critical aviation safety concern, and the detection of FOD has been highly prioritized by aircraft manufacturers. Currently, FOD detection is primarily conducted through manual vision inspection. However, due to the extensive range of FOD types and scales, manual inspection is susceptible to overlooking and misidentifying some small-scale FOD. To handle these difficulties, a lightweight model for detecting small-scale FOD is proposed based on YOLOv5s. This intelligent FOD detection method is of great significance in promoting the aircraft intelligent manufacturing industry. Firstly, we design the backbone feature extraction network (LSGK-GhostNet), which can automatically extract multi-scale features of FOD while reducing computational resources. Secondly, the Deformable Receptive Field Module (DRFB) is proposed by imitating human vision. It utilizes a three-branch structure to integrate various receptive fields for extracting effective FOD scale and contextual information. Additionally, it dynamically adjusts the spatial distribution of the receptive fields. A refined module (DFPN) is designed to amplify the details of small-scale FOD. In addition, the head of the model for detecting the small target is decoupled to strengthen the localizing accuracy for small-scale FOD. Finally, to improve the detection precision of dense detection and address unbalanced sample training, this paper improves the post-processing method and the localization loss of FOD. Detection experiments of FOD are conducted in laboratory and assembly scenarios, The results indicate that our algorithm notably outperforms other methods, and achieves an average accuracy of 94.1%, with a speed of 34.8 Frames per second (FPS) across seven FOD categories. This paper proposes a pioneering way to detect FOD automatically in the assembly process of civil aircraft.

Foreign object debris material recognition based on convolutional neural networks

Material Recognition of Foreign Object Debris using Deep Learning

Foreign Object Debris (FOD) Classification Through Material Recognition Using Deep Convolutional Neural Network With Focus on Metal

Enhancement of Foreign Object Debris Material Recognition using Deep Learning and Machine Learning

Foreign Object Debris Material Recognition based on Ensemble Learning Algorithm

Sppn-Rn101: Spatial Pyramid Pooling Network with Resnet101-Based Foreign Object Debris Detection in Airports

FOD Detection using DenseNet with Focal Loss of Object Samples for Airport Runway

Dynamic Convolution Covariance Network Using Multi-Scale Feature Fusion for Remote Sensing Scene Image Classification

Automatic Detection and Predictive Geolocation of Foreign Object Debris on Airport Runway

Research on FOD Detection for Airport Runway based on YOLOv3

Foreign Object Debris Detection for Optical Imaging Sensors Based on Random Forest

An automatic detecting method for multi-scale foreign object debris in civil aircraft manufacturing and assembly scenario

Deep learning-based foreign object detection method for aviation runways

Towards optimal foreign object debris detection in an airport environment

FOD-A: A Dataset for Foreign Object Debris in Airports

Foreign Object Debris Detection for Airport Pavement Images based on Self-supervised Localization and Vision Transformer

Improved small foreign object debris detection network based on YOLOv5

Automated Detection of Defects with Low Semantic Information in X-ray Images Based on Deep Learning

Foreign Object Debris Detection for Airport Pavement Images Based on Image Random Masking and Semantic Feature Matching

A Sequential CNN Approach for Foreign Object Detection in Hyperspectral Images