论文部分内容阅读
为自动识别椪柑病虫害,研究了以椪柑病虫害为害状多重分形谱特性参数为输入变量的小波神经网络病虫害识别方法。利用改进型分水岭算法提取椪柑病虫害为害状边界,对非连续的边界进行边界跟踪,将过分割区域进行区域合并,标记为害状边界,提取标记区域,生成病虫害为害状目标图像;对病虫害为害状目标图像0°~120°这一主要色相区域4等分,产生4幅色相二值图像;对二值图像进行多重分形分析,计算其标度不变区多重分形谱的高度及宽度;以此高度及宽度作为小波神经网络的输入,进行椪柑病虫害识别,5种病虫害的平均识别正确率为87%。试验结果表明:椪柑病虫害为害状的4对多重分形谱高度及宽度值较充分地反映了椪柑病虫害色相累计信息、分布信息及区间形状的典型特征,能用此方法进行椪柑病虫害机器识别。
In order to identify the pests and diseases of P. citricarpa automatically, the wavelet neural network pest and disease identification method was studied with the input parameters of multiplicative spectral characters of P. citricarpa pests and diseases. The modified watershed algorithm was used to extract the damage-like boundary of P. citricarpa and the boundary of the discontinuous boundary was traced. The overexploited regions were merged into regions and labeled as detrimental boundaries. The main hue region 0 ° -120 ° is divided into 4 equal parts to produce 4 hue binary images. The binary image is analyzed by multifractal analysis to calculate the height and width of the multifractal spectrum of the scale invariant region. With this height And width as the input of wavelet neural network. The correct identification rate of the five kinds of pests and diseases was 87%. The test results showed that the height and width of 4-to-multifractal spectrum of P. citricarpa harmed by P. citricar relatively fully reflected the accumulated information, distribution information and the typical characteristics of the interval shape of P. citricarpa pests, which could be used to identify the plant diseases and insect pests.