基于改進高斯濾波算法的葉片圖像去噪方法

李健 丁小奇 陳光 孫旸 姜楠

摘要：【目的】使用改進的自適應高斯濾波算法對農作物葉片病蟲害圖像進行降噪處理，為葉片病蟲害圖像提供前期預處理的優化手段，從而提高診斷的準確性?！痉椒ā客ㄟ^計算圖像像素矩陣區域內中心點鄰域方差與二維高斯濾波函數的比值，確定高斯標準差，動態生成高斯卷積核，從而形成改進的自適應高斯濾波算法，對病斑圖像進行降噪平滑處理;然后分別模擬不同噪聲強度，比較算法的降噪效果;最后通過峰值信噪比（Peak signal-to-noise ratio，PSNR）定量計算改進前后高斯濾波算法的優化程度。【結果】首先，使用MATLAB 2014b對密刺黃瓜枯萎病斑RGB圖像模擬出3組不同噪聲強度下的干擾場景，并進行歸一化處理;然后，分別利用3種算法對噪聲圖像進行降噪處理，得出當噪聲強度較弱時，改進算法對高斯白噪聲抑制效果明顯;噪聲強度增大時，改進算法的優化程度逐漸下降;其次，分別計算各算法改進前后的PSNR，得出當噪聲強度為0.01、0.02和0.03時，即改進的自適應高斯濾波算法PSNR值分別比傳統高斯濾波提升6.942、6.965和6.718 db;最后，通過計算100組采集葉片圖像降噪處理后的PSNR值，得到改進的自適應高斯濾波的PSNR值平均提高13.8%?！窘ㄗh】采集的農作物葉片圖像試驗材料需廣泛化;推動優化圖像預處理的進程;提升圖像匹配準確性，推動葉片診斷專家系統的研究。

關鍵詞： 圖像降噪;高斯濾波算法;葉片病蟲害;峰值信噪比

中圖分類號： S126;S56 ? ? ? ? ? ? ? ? ? ? ? ? ? ? 文獻標志碼： A 文章編號：2095-1191（2019）06-1385-07

Abstract：【Objective】The improved adaptive Gauss filtering algorithm was used to denoise the image of plant leaf diseases and insect pests， which provided optimization means for the pre-processing of the image of leaf diseases and insect pests， so as to improve the accuracy of diagnosis. 【Method】By calculating the ratio of the variance of the neighborhood of the center point in the image pixel matrix area to the two-dimensional Gauss filtering function， the standard deviation of Gauss was determined， and the Gauss convolution kernel was generated dynamically， thus an improved adaptive Gauss filtering algorithm was formed， and the speckle image was denoised and smoothed. Then， different noise intensities were simulated to compare the denoising effect of the algorithm. Finally， the optimization degree of the improved Gauss filtering algorithm was quantitatively calculated by peak signal-to-noise ratio（PSNR）. 【Result】Firstly， using MATLAB 2014b to simulate three sets of interference scenarios under different noise intensities for RGB image of cucumber fusarium wilt spot， and normalized them. Then， three algorithms were used to denoise the noise image， and it was concluded that when the noise intensity was weak， the improved algorithm could effectively suppress the Gauss white noise. When the sound intensity increased， the optimization degree of the improved algorithm decreased gradually. Secondly， the PSNR of the improved algorithm was calculated before and after the improvement. When the noise intensities were 0.01， 0.02 and 0.03， the PSNR value of the improved adaptive Gauss filtering algorithm was increased by 6.942， 6.965 and 6.718 db， respectively. Finally， the image drop of 100 groups of collected blades was calculated. The PSNR value of the improved algorithm was increased by about 13.8% on average after denoise processing. 【Suggestion】The experimental materials of crop leaf image should be widely used， the process of image preprocessing should be optimized， the accuracy of image matching should be improved， and the expert system of leaf diagnosis should be promoted.

Key words： image noise reduction; Gaussian filtering algorithm; leaf pests and diseases; peak-signal-to-noise ratio

收稿日期：2019-01-21

作者簡介：*為通訊作者，陳光（1961-），博士，教授，博士生導師，主要從事農業信息化及生物信息學研究工作，E-mail：chg61@163.com?！?br>