基于多算法融合的高性能实时图像去雾算法

Research on Dehazing Technology and Image Recognition Based on Multi Algorithm Fusion

摘要:

雾霾等复杂的天气会严重降低自动驾驶汽车采集的图像质量。传统的图像去雾方法存在去雾效果不明显、去雾效率较低的问题，导致汽车环境感知准确度低、去雾实时性差，极大降低汽车环境感知能力, 给汽车自动驾驶带来极大的安全隐患。为解决上述问题，本文引入多尺度空间特征提取和特征融合模块，通过局部连接和权值共享计算优化去雾模型。同时，在去雾模型训练的前向传播和反向传播过程中加入优化的Attention机制，完成基于多算法融合的去雾算法研究。在不同数据集条件下，将本文提出的去雾算法和传统去雾算法的去雾和消融效果进行对比，以峰值信噪比PSNR和结构相似度SSIM作为评价性能的主要指标，分析不同算法的去雾效果。实验结果表明：本文提出的算法去雾效果更明显、去雾效率更高，极大地提高了自动驾驶汽车环境感知能力，从而提高了自动驾驶汽车的行驶安全性能。

Abstract:

Complex weather conditions such as smog and haze can significantly degrade the image quality captured by autonomous vehicles. Traditional image dehazing methods often suffer from limited effectiveness and low efficiency, which leads to reduced accuracy in environmental perception and poor real-time performance. These issues greatly impair the perception capabilities of autonomous vehicles and pose significant safety risks.To address these challenges, this paper introduces a multi-scale spatial feature extraction and feature fusion module to optimize the dehazing model through local connections and weight-sharing computations. Additionally, an optimized attention mechanism is incorporated into both the forward and backward propagation processes during the training of the dehazing model, resulting in a dehazing algorithm based on multi-algorithm fusion.The proposed dehazing algorithm is compared with traditional dehazing methods under various dataset conditions. The evaluation focuses on Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) as the primary performance metrics to analyze the dehazing effectiveness of different algorithms. Experimental results demonstrate that the proposed algorithm achieves more pronounced dehazing effects and higher efficiency, significantly enhancing the environmental perception capabilities of autonomous vehicles and thereby improving their overall driving safety.