Multi-Label Object Detection Network with DenseNet and Data Enhancement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection technologies face challenges in achieving both high detection speed and accuracy, especially in multi-label scenarios where object overlapping and occlusion complicate feature extraction and classification.
Innovation Solution
The method replaces the low-resolution feature layer in the YOLO-V3 network with a densely connected convolutional network (DenseNet) and employs data enhancement techniques like color normalization, brightness transformation, and blurring to improve feature reuse and detection accuracy, enabling real-time detection of multiple object classes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional image processing methods are used for object detection, then detection speed is fast, but detection accuracy is low
Solution Approach 1:
The patent replaces traditional mechanical image processing methods with a deep learning-based object detection network. The network automatically extracts features from input images and performs object detection without manual feature engineering, achieving both high speed and high accuracy by substituting manual mechanical processing with automated neural network inference.
2Measurement precision
If region-based convolutional neural network (R-CNN) is used, then detection accuracy is improved, but real-time detection on high-resolution images is difficult to achieve
Solution Approach 1:
The patent segments the object detection task into multiple stages: first detecting objects in a coarse-grained manner, then refining detections for high-resolution images. This segmentation allows the system to maintain real-time performance on low-resolution inputs while achieving high accuracy on high-resolution images through subsequent refinement processing.
Solution Approach 2:
The patent dynamically adjusts the detection strategy based on image resolution and complexity. For high-resolution images, the system activates enhanced processing paths with higher computational resources, while using lighter processing paths for standard-resolution images, enabling real-time detection across varying conditions.
3Speed
If YOLO network is used, then detection speed is fast, but detection accuracy is slightly lower
Solution Approach 1:
The patent merges the fast single-stage detection approach of YOLO with the high-accuracy multi-stage refinement of R-CNN. The object detection network combines both strategies by first performing rapid one-stage detection and then applying iterative refinement using region-based processing, achieving both speed and accuracy through consolidation of both approaches.
Solution Approach 2:
The patent creates a composite detection system that integrates multiple network architectures and processing strategies. The system combines YOLO-style single-stage detection with R-CNN-style region-based refinement, creating a hybrid architecture that leverages the strengths of both parent approaches to achieve superior overall performance.
4Ease of manufacture
If manual feature extraction is used, then feature extraction is straightforward, but workload is large and universal applicability is lacking
Solution Approach 1:
The patent implements self-service feature extraction through the deep learning network, which automatically learns and extracts features from input images without requiring manual feature engineering. The network performs feature extraction as an integrated part of its detection process, eliminating the need for separate manual feature analysis and reducing overall workload.
Data Source
AI summary
A multi-label object detection method based on an object detection network includes: selecting an image of an object to be detected as an input image; based on a trained object detection network, obtaining a class of the object to be detected, coordinates of a center of the object to be detected, and a length and a width of a detection rectangular box according to the input image; and outputting the class of the object to be detected, the coordinates of the center of the object to be detected, and the length and the width of the detection rectangular box. The method of the present invention can perform real-time and accurate object detection on different classes of objects with improved detection speed and accuracy, and can solve the problem of object overlapping and occlusion during the object detection.