基于图像频谱特征的稻飞虱识别方法

为准确、快速的识别稻飞虱种类,采用自行设计的野外环境昆虫图像采集装置获取稻飞虱背部图像,通过对背景与目标像素的统计,选取140为阈值,对稻飞虱图像的蓝色通道进行二值化,将背景与目标分割开,采用形态学滤波以及开运算,与灰度图像进行与操作,获取单个稻飞虱虫体背部区域的灰度图像。然后对灰度图像进行二维傅里叶变换,获得虫体背部图像的二维傅里叶频谱。最后以ll(l=1,2,…,6)的二维频谱窗口数据作为稻飞虱特征参数,建立Fisher判别函数。训练集和验证集的试验结果表明,选用33二维傅里叶频谱窗口数据建立的判别模型,稻飞虱正确识别率可达到90%以上。该方法可以实现田间稻飞虱的自动识别。

dentification method for rice plant hoppers based on image spectral characteristics

To accurately and rapidly identify the rice plant hoppers, a novel method for identification of the rice plant hoppers was proposed by using image processing and image spectra analysis. At first, the automatic digital image acquisition device developed by us was used to capture the hopper images. The threshold 140 was utilized for the segmentation of the background and the insects after statistic and analyzing gray values of the pixels in the background and the insects. Then morphological filting, opening and AND operations were conducted on the insect images and the back images of each insect were obtained. Two-dimensional Fourier spectra for back images of the insects were extracted through the Fourier transform. Finally, the fisher discriminant function was established based on the two-dimensional spectrum data and was used for the identification of the rice plant hoppers. The training and validation results showed that the correct recognition rates for the rice plant hoppers were more than 90%, indicating that the proposed method has potential to automatically identify the rice plant hopper.