【作者】 林妙玲；

【导师】 李建平；

【作者基本信息】 浙江大学 ， 农业电气化与自动化， 2007， 硕士

【摘要】 虾类是主要的水产品，虾的加工处理包括剥壳、去头、分级和包装等。目前市场上有虾剥壳的设备，但是虾去头绝大部分还是由人工手工来完成的。虾的头部识别在新鲜虾去头、剥壳虾去头、分级等虾加工处理中具有重要意义。本文利用机器视觉技术和小波变换低通滤波研究虾的图像处理和虾头部识别。主要内容和结果如下：1．综述了虾剥壳和去头的研究应用情况，特别阐述了机器视觉在水产品中的研究现状，例如在虾体特征识别、分级、品质检测和包装上的研究情况。2．研究究分了虾图像的预处理方法，包括中值滤波、迭代阈值分割、形态学开闭运算、轮廓提取等。3．分析对比发现3次小波变换低通滤波处理后的图像在保持虾基本轮廓的前提下较大限度地保留了轮廓曲线上虾头连接点的细节。利用链码表示虾的背部和腹部曲线，通过计算链码上水平和垂直方向具有最大阶跃变化的码值点的坐标，即可识别虾轮廓曲线上头部和身体连接点的坐标。整个识别过程中的内容和结果有：1)链码上虾头尾端点的识别算法。2)根据小波滤波后的虾体细化图像得到的头部端点和尾部端点形成的直线的斜率．K的不同建立了10种位姿模式和判断方法。3)判断出虾体的位姿后，以一定的算法来识别背部和腹部链码。4)以背部链码上的头部端点为起始点，以特定的算法识别得到虾头和身体连接点处的两个坐标。5)试验结果表明：基于小波变换低通滤波的虾背部和腹部连接点坐标的识别误差约为3.53mm。当误差范闱在4mm以内时，第一批冰冻解冻虾识别算法准确率约为91.6％；第一批非冰冻虾识别算法的准确率约为85％。4．归纳了同批次长度不均匀虾的虾头和身体连接处的识别方法中需要调整的参数及参数计算公式。。5．结合虾头和身体连接处的颜色差异，提出增进算法识别精度和效率的研究设想。更多还原

【Abstract】 Shrimp is one of the main products in world aquatic product volume of trade, shrimp’s processing includes peeling, dehead, grade and package. Nowadays there are shrimp peeling devices in market, but the operation of shrimp dehead is mostly done by manual work. Identification of shrimp head has signicificant meaning in shrimp processing. This paper studys shrimp image process and shrimp head identification by using machine vision and wavelet transform lowpass filtering. The research content and rusults are as following:1. Introduce the research and application of shrimp peeling and dehead, especially machine vision research situation in aquatic p(?)oduct, such as shrimp characteristic recognition, grade, quality detection and package.2. Study shrimp image pretreat method, including median filtering, iterated threshold segmentation, morphology open and close operation and contour extraction..3. Wavelet transform lowpass filtering is used to analyse image contour. Compare with experiment results, it is found that three times wavelet transform lowpass filtering is able to preserve details of shrimp contour while keeping shrimp basic contour. Chain code is used to represent shrimp back and abdomen curve, the point coordinates which connect head and body can be identified by calculating the coordinates which has the maximum change at horizontal and vertical orientation in chain code. The content and results of whole recognition process are shown below:1) Shrimp head and tail points identification algorithm is given.2) Building and judgment of 10 shrimp postion model. Different shrimp positions are divided according to straight line slope. The straight line is formed by head and tail points in shrimp wavelet filtering thinning image.3) Identification of back and abdomen chain code. A certain algorithm is proposed to identify back and abdomen chain code when knowing shrimp position.4) Identification of shrimp head and body connecting points in chain code. Start searching along chain code from shrimp head point in back chain code, find shrimp head and body connecting segment vector by using specific algorithm.5) Results show that coordinates error of shrimp back and abdomen connecting points based on wavelet lowpass filtering is about 3.53mm. When the distance error under 4mm, the accuracy of first frozen and unfrozen shrimp identification algorithm is 91.6%, while the first unfrozen shrimp is 85%.4. Identification meathod of shrimp head and body connecting points when shrimp’s length is nonuniform in the same batch is concluded. Parameters and their calculating formulas are given. 5. Based on the head and body connecting points colour difference, further research idea to improve Identification algorithm accuracy and efficiency is put forward.更多还原