番茄缺素叶片的图像特征提取和优化选择研究

在基于计算机视觉技术对无土栽培番茄营养元素缺乏智能识别研究中,对不易被肉眼判别的缺氮和缺钾初期叶片进行图像特征的提取和优化选择研究，以提高识别的准确率。提出了应用相对差值百分率直方图提取缺素叶片的颜色特征，应用差分算子提取纹理的时域特征、应用傅里叶变换提取纹理的频域特征、应用小波包提取纹理的时频特征等的新方法，并新提出从颜色和纹理时域、频域、时频域等多个角度集成提取缺素叶片图像的有效特征，利用遗传算法对提取的众多特征项进行优化选择，以使诊断识别用的信息分类能力最优。试验表明，该方法识别的准确率较高，达到95%～92.5%，而且可以比肉眼识别提前6～10 d。

Extracting and selecting features of leaf images for diagnosing nutrient deficiency diseases in tomatoes

Diagnosing nutrient deficiency diseases of tomatoes in soilless agriculture was studied. Image features that were not recognized by human eyes easily from nutrient deficient (N,K) tomato leaves were extracted and selected, so that diagnosing accuracy was improved. The color features of nutrient deficient tomato leaves were extracted by the percent color histogram method. The texture features of the leaves were extracted by the percent differential histogram method. The texture features of the frequency region were extracted by the Fourier Transform method. The texture features of the time frequency region were extracted by the wavelet packet method. Application of genetic algorithm to select the above mentioned features was studied in order to get the best information for diagnosing. The experimental results show that: the accuracy of this diagnosis system is higher, its accuracy reaches 95%~92%, and it can accurately diagnose diseases 6~10 d before their symptoms can be recognized by human eyes, and thus reduce damage to the agricultural products.