基于近邻法聚类和改进Hough算法的猪胴体背膘厚度检测

为了解决猪胴体背膘厚度在人工测量中准确率、效率低以及存在对样本造成污染的问题,该文基于计算机视觉和图像处理技术提出一种检测背膘厚度的算法。算法主要分成背膘部分检测和测量部位的直线检测。前者通过图像分割、特征点的检测以及漫水填充等方法实现,能准确提取猪胴体背膘部分。后者在图像预处理后,首先通过感兴趣区域(region of interest,ROI)提取猪胴体肋排区域;然后利用设定好的浮动窗口进行全幅图像的扫描,通过平滑后的平均灰度线特征提取肋骨的目标像素点;最后,基于近邻法利用目标像素点间的邻近关系对其进行聚类,找到胴体第6、7根肋骨,并采用基于已知点的Hough变换提取测量直线,将测量部位的直线映射到背膘部分,则可实现对猪胴体背膘厚度准确测量。试验结果表明,在对背膘厚度测量误差小于2 mm时,检测准确率可达92.31%,该文提出的方法能对猪胴体背膘厚度的测量位置进行准确定位和测量。

Detection of pork backfat thickness based on nearest neighbor clustering and improved Hough algorithm

Abstract: The backfat thickness of pig carcasses refers to the fat thickness between the sixth rib and seventh rib, and the fat thickness is uniform, and is the thickest in this position. Backfat thickness is one of the most important indexes to assess pork quality. However detection of backfat thickness in most of the pork slaughtering houses and enterprises depends on trainees using measuring tools or making judgments directly by eye in China. This manual detecting method demands not only great labor but also lacks objectivity and accuracy. Additionally, hand contact can cause great adulteration to the pig samples. The objective of this research was to investigate a method for detecting backfat thickness based on computer vision and image processing technologies. The pig carcass images were collected by a machine vision image acquisition system on the slaughter line. The system consisted of an image acquisition device, light, a single chip microcomputer, a detection control button, and a computer and image processing algorithm equipped into the self developed system software. A black background plate was placed behind the pig carcass in order to adapt to the complexity of the environment. When a half carcass reached the camera view, the operator pressed the control button to acquire images of the carcass. And these collected images were automatically stored in the computer for further image processing. The algorithm consisted of two parts: the detection of the backfat part and the location. Some methods such as image segmentation, feature point detection, and flood fill were adapted to extract the backfat part. The method of determining the measurement position was as follows. First, the region of interest (ROI) was obtained. In this step, the rib area was extracted from the pig carcass. Then the floating window was used to scan the whole ROI image. The size of the scanning window was 20×1 and the direction of scanning was from top to bottom in each line of image. The average gray values in each scanning widow were calculated to obtain the distribution of the average gray value in each column. The feature points of the ribs were extracted by the characteristics of the average gray level line on each column of the ROI image. Next, points on the sixth and seventh ribs were clustered based on a nearest neighbor clustering algorithm. The points of each column were averaged, and they became new feature points between the sixth and seventh ribs. The horizontal and vertical coordinates of the known point were the average of new feature points. At last, we extracted the measuring line based on Passing a Known Point Hough Transform (PKPHT). The slope between two points, which belonged to the same line, was calculated and the slope accumulator was voted. The peak of the slope accumulator corresponded to the slope of the line to detect. Backfat thickness can be measured accurately by mapping the line to the backfat part. The results showed that this method can nondestructively determine the measurement position and measure the backfat thickness of pig carcasses after validation on multiple samples. And this method is of great significance for the development of an automatic classification system.