黄曲条跳甲图像识别的综合实验设计  

Design on comprehensive experimental of image recognition for striped flea beetle(Phyllotreta striolata)

作　　者：岑冠军[1] 房少梅[1] 杨德贵[1] 张胜祥[1] 张连宽[1] CEN Guanjun;FANG Shaomei;YANG Degui;ZHANG Shengxiang;ZHANG Liankuan(College of Mathematics and Informatics,South China Agricultural University,Guangzhou 510642,China)

机构地区：[1]华南农业大学数学与信息学院,广东广州510642

出　　处：《实验技术与管理》2021年第10期198-203,210,共7页Experimental Technology and Management

基　　金：2019年广东省高等教育教学研究和改革项目(2019-102);广东省重点领域研发计划项目(2019B020217003)。

摘　　要：设计了识别田间黄色害虫诱捕板图像上黄曲条跳甲的综合实验项目。对原始图像在HSV颜色空间下采用OTSU算法去除背景后,通过设定连通域面积阈值初筛出候选目标,构建椭圆模型拟合候选目标的轮廓,提取椭圆区域的形态学、颜色与纹理等特征,根据特征应用浅层神经网络对候选目标进行识别。上述算法采用并行计算的MATLAB程序实现,具有良好的准确性、稳健性和泛化能力。该实验设计涵盖了农业害虫监测、图像处理、解析几何、机器学习和并行计算等知识,具有综合性、应用性、交叉性和探索性。将其用于数学建模课程的教学,有助于提高学生的创新和实践能力。A comprehensive mathematical experiment was designed to identify striped flea beetles on images of yellow sticky traps in the field.Firstly,the backgrounds of original images were removed by OTSU algorithm in HSV color space.Secondly,candidate targets were screened by setting the connected domain area threshold.Thirdly,the ellipse model was constructed to fit the contour of the candidate targets,and the morphological,color and texture features of the ellipse region were extracted.Finally,the shallow neural network has been used to recognize the candidate targets with the above features.The algorithm has been implemented by parallel computing MATLAB program,and performed well in accuracy,robustness and generalization ability.The experimental design covered the knowledge of agricultural pest monitoring,image processing,analytic geometry,machine learning and parallel computing.Hence,the experiment design was provided with comprehensiveness,applicability,crossover and exploratory.Applying it to the teaching of mathematical modeling course is beneficial to improve students'innovation and practical ability.

关 键 词：综合实验设计 黄曲条跳甲 图像识别 数学实验 

分 类 号：G642.0[文化科学—高等教育学] O29[文化科学—教育学]