Road image small target detection system of lightweight cross-layer feature pyramid network

Abstract

The invention relates to the technical field of computer vision, and particularly discloses a road image small target detection system of a lightweight cross-layer feature pyramid network. The system comprises a feature extraction module, a lightweight cross-layer attention module, a detection head module and a model compression module. The feature extraction module adopts a lightweight backbone network to extract a multi-scale feature map; the lightweight cross-layer attention module enhances feature fusion through a cross-layer channel attention and space attention mechanism; the detection head module adopts a lightweight convolutional layer and a classification-regression branch to carry out multi-scale prediction; the model compression module optimizes the model through knowledge distillation and channel pruning. The method effectively reduces the calculation complexity and parameter quantity, improves the detection precision and efficiency of the small target in the road image, and is suitable for embedded equipment and real-time scenes.

Images