杨信廷, 刘蒙蒙, 许建平, 赵丽, 魏书军, 李文勇, 陈梅香, 陈明, 李明. 自动监测装置用温室粉虱和蓟马成虫图像分割识别算法[J]. 农业工程学报, 2018, 34(1): 164-170. DOI: 10.11975/j.issn.1002-6819.2018.01.022
    引用本文: 杨信廷, 刘蒙蒙, 许建平, 赵丽, 魏书军, 李文勇, 陈梅香, 陈明, 李明. 自动监测装置用温室粉虱和蓟马成虫图像分割识别算法[J]. 农业工程学报, 2018, 34(1): 164-170. DOI: 10.11975/j.issn.1002-6819.2018.01.022
    Yang Xinting, Liu Mengmeng, Xu Jianping, Zhao Li, Wei Shujun, Li Wenyong, Chen Meixiang, Chen Ming, Li Ming. Image segmentation and recognition algorithm of greenhouse whitefly and thrip adults for automatic monitoring device[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(1): 164-170. DOI: 10.11975/j.issn.1002-6819.2018.01.022
    Citation: Yang Xinting, Liu Mengmeng, Xu Jianping, Zhao Li, Wei Shujun, Li Wenyong, Chen Meixiang, Chen Ming, Li Ming. Image segmentation and recognition algorithm of greenhouse whitefly and thrip adults for automatic monitoring device[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(1): 164-170. DOI: 10.11975/j.issn.1002-6819.2018.01.022

    自动监测装置用温室粉虱和蓟马成虫图像分割识别算法

    Image segmentation and recognition algorithm of greenhouse whitefly and thrip adults for automatic monitoring device

    • 摘要: 为了监测温室黄瓜作物虫害种类、数量变化情况以预测虫害发展趋势,该文以粉虱和蓟马为例,提出了一种基于Prewitt、Canny边缘检测算子分割和SVM(support vector machine)的温室粉虱和蓟马诱虫板的图像识别算法。该方法利用HSI(Hue-Saturation-Intensity)颜色空间的I分量与L*a*b*颜色空间的b分量二值图像中害虫目标与背景的高对比性,再分别相应地利用Prewitt算子和Canny算子进行单头害虫边缘分割,再经过形态学处理,最后融合这两幅二值图像完成单头害虫区域的提取。然后提取害虫的5个形态特征(面积、相对面积、周长、复杂度、占空比)及9个颜色特征(Hue-Saturation-Value颜色空间、HSI颜色空间、L*a*b*颜色空间各分量的一阶矩),并对这14个特征参数进行归一化处理,将特征值作为SVM的输入向量,进行温室粉虱和蓟马的诱虫板图像识别。通过分析比较不同向量组合的BP(back propagation)与SVM的害虫识别率、4种不同SVM核函数的害虫识别率,发现颜色特征向量是粉虱和蓟马识别的主成分,且SVM的识别效果优于BP神经网络、线性核函数的SVM分类性能最好且稳定。结果表明:平均识别准确率达到了93.5%,粉虱和蓟马成虫的识别率分别是96.0%和91.0%,能够实现温室害虫的诱虫板图像识别。该研究可以为虫害的监测与预警提供支持,为及时采取正确的防治措施提供重要的理论依据。

       

      Abstract: Abstract: Pests were one of the important factors for crop loss, so accurate counting and identification of insect were important for pest detection and preventive measures. To monitor the variety and quantity of cucumber pests in greenhouse and predict the development trend of pest, an image recognition algorithm based on Prewitt, Canny edge detection operator segmentation and support vector machine (SVM) was proposed for greenhouse whitefly and thrip on the trap board. In recent years, due to the improvement of computer hardware and imaging equipment, automatic recognition of pests based on image processing technology has been widely studied. The common counting and recognition methods were mainly based on computer vision, the pest target was segmented from the background image firstly, and then the pests were counted and identified. Classification and counting inevitably required precision control. Firstly, pest images were captured by field automatic pest monitoring camera equipment. It was necessary to divide the pest area from the image in order to extract the valid feature parameters. For the sake of raising the effect of edge detection, it was significant to strengthen the contrast between the target and the background. The method used the high contrast of the pest target and the background in the b component binary image of the L*a*b* color space and the I component binary image of the HSI (hue-saturation-intensity) color space, and used the Prewitt operator and the Canny operator to perform the single-headed pests edge segmentation respectively. After morphological processing, differential method was used to remove the interference and tiny hole filled, and the 2 binary images were merged to complete the extraction of single-headed pests. The method extracted 5 morphological features (area, relative area, perimeter, complexity, and duty ratio) of pests, and because of the great diversity of color in pests, 9 color features (the first moment of each component of HSV (hue-saturation-value) color space, HSI color space and L*a*b* color space) were extracted. And the 14 characteristic parameters were normalized, and used as the input vector of the SVM to identify the image of cucumber pests. The pest identification rates of BP (back propagation) and SVM were compared, and the pest identification rates of 4 SVM kernel functions were also compared. The results show that the color eigenvector is the principal component of pest identification and SVM is better than BP neural network. SVM classification of linear kernel function has the best classification performance and stability, the average recognition accuracy is 93.5%, and the pest identification rates of whitefly and thrip are 96.0% and 91.0% respectively. The proposed method can realize the image recognition of greenhouse sticky traps. The study can support the pest detection and early warning, and provide important basis for timely and correct control measures.

       

    /

    返回文章
    返回