APP下载

基于DCP算法增强暗纹东方鲀胚胎图像

2021-01-29冯国富,刘亚蕊,陈明,翁正,王耀辉

江苏农业学报 2021年6期

冯国富,刘亚蕊,陈明,翁正,王耀辉

摘要:为了能够精准地识别出暗纹东方鲀胚胎发育的各个时期,提高人工繁殖和杂交育种的成活率,采用暗通道先验(DCP)算法和它的2个反演以及推导,对胚胎图像进行增强。推导出4种透射率,再结合像素值的缩小和放大最终生成8种增强效果。其中图像增亮与图像增暗结合的A+X组合算法效果最佳,增强后的图像细节信息更加丰富,各个时期的特征更加清晰。选用多目标检测算法YOLOv5网络对图像中的胚胎进行提取和分类。结果表明,增强后图像分类结果的准确率比原始图像提高4.5%,损失函数降低0.16%。

关键词:暗纹东方鲀;胚胎检测;暗通道先验;图像增强;YOLOv5网络

中图分类号:S961.2文献标识码:A文章编号:1000-4440(2021)06-1493-08

Image enhancement of Takifugu obscurus embryos based on DCP algorithm

FENG Guo-fu1,2,LIU Ya-rui1,2,CHEN Ming1,2,WENG Zheng1,2,WANG Yao-hui3

(1. School of Information, Shanghai Ocean University, Shanghai 201306, China;2.Shanghai Ocean University, Key Laboratory of Fishery Information, Ministry of Agriculture and Rural Affairs, Shanghai 201306, China;3.Nantong Longyang Aquatic Products Co., Ltd., Nantong 226634, China)

Abstract:In order to accurately identify each stage of embryonic development of Takifugu obscurus and improve the survival rate of artificial reproduction and cross breeding, dark channel prior (DCP) algorithm and its two inversions and derivation were used to enhance embryo image. Among them, four kinds of transmittances were derived in this study, and eight kinds of enhancement effects were finally generated by combining the reduction and amplification of pixel values. Through comparative analysis, A+X combination algorithm combining image brightening and image darkening had the best effect. The enhanced image details were more abundant, and the features of each period were clearer. The YOLOv5 network, a multi-target detection algorithm, was selected to extract and classify embryos in the images. The results showed that the classification accuracy of enhanced image was 4.5% higher than that of original image, and the loss function was 0.16% lower.

Key words:Takifugu obscurus;embryos detection;dark channel prior;image enhancement;YOLOv5 network

暗紋东方鲀与鲥鱼、刀鱼并称“长江三鲜”,享誉古今中外。近几年,暗纹东方鲀仅依靠自然繁殖已不能满足市场的需求,对它进行人工繁殖逐渐成为研究热点[1]。在人工繁殖和杂交育种过程中对胚胎发育的各个时期进行检测与分类显得非常重要[2-3]。通过观察胚胎的形态特征识别发育时期,选择适宜的温度、盐度等环境培育胚胎,来提高胚胎的孵化率[4-5]。然而,人工肉眼识别胚胎发育的各个时期效率较低且准确性不高。因此,我们使用基于图像处理的人工智能技术来解决这一难题,当前,国内外将人工智能技术应用在胚胎检测、分类方面的研究几乎是空白。

胚胎图像通过数字显微镜获取,由于胚胎带有水珠等原因,导致图像存在光照不均、细节不清晰、色彩对比度不明显等问题,因此需要对获取的胚胎图像进行特征增强。……

登录APP查看全文