摘要
针对蚕茧分选中传统人工分类不准确、效率低的问题,提出一种基于改进卷积神经网络与图像处理结合的蚕茧识别方法。该方法模型共5层卷积层、5层池化层、4层全连接层,针对蚕茧表面受茧丝缠绕导致纹理不清晰以及光照不均问题,利用主成分分析法和颜色空间转换法对蚕茧图像进行特征增强。试验得该模型识别准确率达96%,其中经图像处理后的柴印茧识别准确率从86.21%提高到93.55%,色斑茧识别准确率从96%提升到100%。试验表明该方法可有效改善蚕茧识别效果,该模型可满足蚕茧识别分类需求,为实际生产提供有效参考依据。
In view of the inaccuracies and low efficiency of traditional manual classification in cocoon sorting,a cocoon recognition method based on an improved convolutional neural network and image processing was proposed.The method s model has 5 convolutional layers,5 pooling layers,and 4 full connection layers.The method uses principal component analysis(PCA)and color space conversion to enhance the characteristics of the cocoons images.The experimental results show that the identification accuracy of the model is 96%,with the identification accuracy of Chaiyin cocoons increasing from 86.21%to 93.55%,and the identification accuracy of colored cocoons increasing from 96%to 100%.The results show that the image processing method can significantly improve the identification effect.Furthermore,the model meets the cocoon classification needs and provides an effective reference for actual production.
作者
周显沁
韩震宇
刘成源
Zhou Xianqin;Han Zhenyu;Liu Chengyuan(School of Mechanical Engineering,Sichuan University,Chengdu,610000,China)
出处
《中国农机化学报》
北大核心
2023年第5期100-106,120,共8页
Journal of Chinese Agricultural Mechanization
基金
重庆市璧山区2021农业科研项目(Bskj20210011)。
关键词
图像处理
卷积神经网络
主成分分析
蚕茧分选
image processing
convolution neural network
principal component analysis
cocoon sorting
