基于随机森林的鼠类头骨自动识别系统开发

Development of rat skull automatic identification system based on random forests

目的以实现鼠类头骨自动鉴定为目的,设计并开发了基于随机森林的鼠类自动识别系统。方法系统应用计算机图像识别技术,使用计算机视觉库OpenCV和随机森林算法,设计鼠类头骨识别的训练和识别程序,训练模块包括图像输入、图像预处理、特征提取、训练分类器和分类器数据储存5个部分,识别模块的流程又包括图像输入、图像预处理、特征提取、模式识别和返回识别结果5个部分。通过自动提取鼠类头骨上颌正面图像的32个数学形态学特征(如偏心率、叶状性等),对啮齿目4科13种鼠进行自动识别。结果实际运行结果显示,系统的平均识别率>80%,且每鼠种的识别正确率均在70%以上,说明该系统操作性强,鉴定结果比较可靠。结论借助计算机技术可以实现鼠类的自动识别,但目前该系统还处于研究的初步阶段,如何进一步提高识别率尚需更多的研究。

Objective Rodents can survive in any existing habitat, and are the largest order of mammals in terms of both variety and quantity of species. Unlike other mammals, their classification can be variable and complex, due to the large number of species. Furthermore, the inter-species characteristics and traits tend to be convergent. They pose a seriously increasing threat to agriculture. Hence it is crucial to find effective measures to manage and control rodent infestation, which requires confirmation of their taxonomic statuses precisely and quickly. However, an expert on one species or family may be unfamiliar with another. These issues have increased the demand for digitized software tools that can recognize and characterize rodent skulls from images. In this study we developed a system, named ＂Rodents Skull Automatic Identification System＂, based on random forests. Methods The training module and recognition module of the system were designed based on the recognition technology of computer science, OpenCV, and random forests. The training module included image input, image preprocessing, feature extraction, pattern recognition, and identification result return. The recognition module included image input, image preprocessing, feature extraction, training classifier, and classifier data storage. The system identified rodents of 13 species from 4 families through automatic extraction and analysis of 32 mathematical morphological features on dorsal maxillary images, such as eccentricity and compactness. Results The system could identify 13 species among 4 families of rodents. The results showed that the average identification accuracy rate was above 80%, and the identification accuracy rate of each species was above 70%, indicating that the system was highly reliable in recognition of rodents. Conclusion Rodents can be automatically identified with the aid of computer technology. However, this system is only a preliminary study, and it requires further studies to improve the recognition rate.