摘要
针对鱼类识别面临着光照强度、各背景栖息地的变化和不同物种在视觉上具有相似性等方面的问题,提出一种新的基于多特征相结合及粒子群优化SVM的鱼类分类方法。该方法采用在原始图像中提取颜色、方向梯度直方图(HOG)和灰度共生矩阵(GLCM)特征构成特征向量,并提出选择设置最佳权重比的方法进行特征融合,采用PCA技术对提取的特征向量进行降维,以消除冗余数据。结果表明,该方法在实际采集的数据集上的准确率达94.7%,同属类鱼识别最高准确率93.75%,该方法可以应用于实际的鱼类图像数据集,实现对鱼类生物多样性的有效监测。
This effort faces problems in terms of light intensity changes,background habitats,and visual similarities among different species.This paper proposes a new fish classification method based on multi-feature combination and a particle swarm optimization SVM.Colour,HOG and GLCM features are extracted from the original image to form a feature vector.And,the method of selecting the optimal weight ratio is presented to fuse the features.To reduce elimination,a PCA algorithm is applied to reduce the dimensionality of the feature matrix.Finally,the paper proposes a classification algorithm in which a particle swarm optimization SVM can be used for fish recognition.Experimental results show that the accuracy of the proposed method in the real-world fish image dataset is 94.7%.In addition,the highest accuracy rate of 93.75%was obtained from the recognition of fish of the same genus in this paper.The methods can thus be applied to a real-world fish image dataset to monitor fish biodiversity inexpensively and effectively.
作者
丁顺荣
肖珂
Ding Shunrong;Xiao Ke(College of Information Science&Technology,Hebei Agricultural University,Baoding,071000,China)
出处
《中国农机化学报》
北大核心
2020年第11期113-118,170,共7页
Journal of Chinese Agricultural Mechanization
基金
国家自然科学基金(31801782)。
关键词
鱼类识别
特征融合
粒子群优化
支持向量机
fish identification
feature fusion
particle swarm optimization
SVM
