论文部分内容阅读
[目的]为克服光照不均引起的低对比度、反光、阴影、光斑及遮挡等对大田复杂背景下小麦冠层图像分割的干扰。[方法]本文设计了一种结合脉冲耦合神经网络(Pulse Coupled Neural network,PCNN)与同态滤波的自适应图像增强和基于L*a*b*颜色空间α角度模型的K均值聚类分割算法。首先,将小麦冠层图像转换到HSI颜色空间,采用自适应算法对HSI空间的I分量进行增强处理,适当调节饱和度S分量,补偿光照强度分布不均,去除阴影及拉大对比度;其次,将增强处理后的图像映射到L*a*b*颜色空间,提取a*、b*分量建立α角度模型;最后,基于α进行K均值聚类分割处理。[结果]拔节前后光强不一、光照不均的冬小麦冠层图像的分割试验结果表明,本文算法可一定程度避免基于L*a*b*颜色空间α角度分量K均值聚类的过分割现象;改善基于HSI空间H分量K均值聚类的欠分割缺陷,且对光斑、阴影遮挡、反光突出的图像分割更完整、准确。[结论]可为大田复杂背景下光照多变的作物冠层图像分割提供参考方法。
[Objective] The research aimed to overcome the interference of low contrast, reflectivity, shading, spot and occlusion on wheat canopy image segmentation under field complicated background. [Method] This paper designs a K-means clustering algorithm based on Adaptive Coupling Neural Network (PCNN) and Homomorphic Filtering and K-means clustering algorithm based on L * a * b * color space . Firstly, the wheat canopy image is transformed into the HSI color space, the I component of the HSI space is enhanced by an adaptive algorithm, the S component of the saturation is adjusted appropriately to compensate the uneven distribution of light intensity, remove the shadow and enlarge the contrast; second, The enhanced image is mapped into the L * a * b * color space, and the a * and b * components are extracted to establish the alpha angle model. Finally, the K-means clustering segmentation is performed based on α. [Result] The segmentation results of canopy images of winter wheat with different light intensity and uneven illumination before and after jointing show that this algorithm can avoid the over - segmentation based on K - means clustering of α - angle component in L * a * b * color space to some extent ; To improve the under-segmentation defect based on H component K-means clustering in HSI space, and more complete and accurate image segmentation for spot, shadow occlusion and reflective highlighting. [Conclusion] This method could provide a reference method for crop canopy image segmentation with varied illumination in complex fields.