Multi-Scale Feature Fusion Deep Learning Model for Small Vehicle Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection algorithms, such as YOLOv8, struggle to quickly and accurately recognize and position small-size vehicles due to limitations in feature extraction and processing efficiency.
Innovation Solution
A small-size vehicle detection deep learning model incorporating a C2f_DCNv3 module, SPPF_LSKA module, C2f_SCConv module, and MSK_Detect module for enhanced feature extraction and prediction, optimizing the Backbone, Neck, and Head networks to improve detection accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional object detection algorithms (YOLOv8) are used, then processing speed is improved, but detection accuracy for small-size vehicles deteriorates
Solution Approach 1:
The detection task is segmented into multiple stages with different detection heads. The first detection head processes standard-sized objects, while the second detection head specifically processes small-size vehicle objects. This segmentation allows each head to be optimized for its specific task, improving both speed and accuracy for small vehicles without compromising overall processing efficiency.
Solution Approach 2:
The patent applies local quality by introducing a specific small object detection head with specialized feature extraction modules (such as the LSKA attention mechanism and multi-scale feature fusion) that are tailored for small-size vehicles. This localized optimization ensures that small vehicles receive enhanced processing attention while maintaining the efficiency of the overall detection system.
2Measurement precision
If feature extraction depth is increased to improve small object detection, then detection accuracy improves, but calculation and storage costs increase
Solution Approach 1:
The patent performs preliminary action by introducing an attention mechanism (LSKA) and multi-scale feature fusion modules in the feature extraction process. These modules pre-process and prioritize important features early in the network, allowing subsequent layers to work with already-refined feature representations. This reduces the computational burden in deeper layers while maintaining high detection accuracy.
Solution Approach 2:
The patent addresses feature extraction efficiency by operating in another dimension through multi-scale feature fusion. Instead of simply increasing network depth, the model fuses features from multiple scales and dimensions, capturing small vehicle information more effectively without proportionally increasing computational costs. This dimensional approach allows efficient extraction of small object features.
Data Source
AI summary
A small-size vehicle detection deep learning model based on feature fusion of multi-scale modules is provided, which solves the problem of small-size vehicle image detection. The model includes a Backbone network, a Neck layer and a Head network, wherein a C2f_DCNv3 module based on the combination of deformable convolution v3 (DCNv3) and a cross stage feature fusion (C2f) module and an SPPF_LSKA module based on the combination of a spatial pyramid pooling fast (SPPF) layer and a large separable kernel attention (LSKA) module are introduced into the Backbone network; a C2f_SCConv module based on the combination of spatial and channel reconstruction convolution (SCConv) and a C2f module is introduced into the Neck layer; and a multi-scale kernel detection (MSK_Detect) module is introduced into the Head network.