SMFN: A small target detection method in nylon yarn package images based on neural network

Nylon, a vital synthetic fiber, plays a crucial role in various industries. Detecting small-sized defects on nylon yarn packages presents a formidable challenge in quality control. In response to this challenge, we propose an innovative approach based on YOLOv5, introducing the Spatially Sensitive Multi-Mode Feature Fusion Neural Network (SMFN). The key innovation in SMFN is the development of a spatially sensitive attention module (SSA). This module enhances feature extraction, capturing a wealth of spatial information while improving the detection accuracy for smaller targets. Furthermore, we introduce a multi-mode feature fusion module (MFF), replacing upsampling module. This modification enables the neural network to autonomously learn feature fusion, leading to superior performance. On a dedicated dataset of nylon yarn packages, The SMFN model achieves a remarkable [email protected] score of 66.9%. This represents a substantial 5.7% improvement over the baseline YOLOv5 model, while maintaining a rapid detection speed of 59.42 frames per second. In summary, SMFN empowers the swift and high-precision small target detection of defects in nylon yarn packages, promising enhanced quality control in manufacturing processes.