YOLOv3-Tiny Object Detection Model Small Target Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The YOLOv3-Tiny model struggles with detecting small targets due to hardware limitations and occlusions, leading to high false detection and missed detection rates, which are not adequately addressed by existing object detection algorithms.
Innovation Solution
The proposed target object detection model improves upon YOLOv3-Tiny by merging low-level and high-level feature maps through concatenating layers, allowing the model to retain low-level information and enhance the detection of small objects by reducing the filtering effect of convolutional and pooling layers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If YOLOv3-Tiny model is used for real-time detection, then detection speed is improved, but detection accuracy of small targets deteriorates
Solution Approach 1:
The patent merges low-level feature maps (from convolutional layers) and high-level feature maps (from pooling layers) through concatenating layers. This combination allows the model to retain detailed low-level information while incorporating high-level semantic information, thereby improving small target detection accuracy without sacrificing the real-time detection capability of the YOLOv3-Tiny architecture.
2Productivity
If convolutional and pooling layers are used for feature extraction, then feature processing efficiency is improved, but low-level information is lost due to filtering
Solution Approach 1:
The patent extracts low-level feature information from convolutional layers that would otherwise be filtered out by subsequent pooling operations. By introducing concatenating layers that preserve and integrate these low-level features with high-level features, the model prevents information loss while maintaining efficient feature processing through the standard convolutional and pooling layer architecture.
Data Source
AI summary
A target object detection model is provided. The target object detection model includes a YOLOv3-Tiny model. Through the target object detection model, low-level information in the YOLOv3-Tiny sub-model can be merged with high-level information therein, so as to fuse the low-level information and the high-level information. Since the low-level information can be further used, the comprehensiveness of target detection is effectively improved, and the detection effect of small targets is improved.