孟令濤 孫國(guó)強(qiáng)



摘 ?要: 藥用植物目標(biāo)檢測(cè)可以有效應(yīng)用于藥用植物的圖像識(shí)別以及圖像的語義分割。文章對(duì)已有的算法做了優(yōu)化,使其對(duì)自然環(huán)境下的藥用植物目標(biāo)檢測(cè)更為準(zhǔn)確,提出了一種基于選擇性搜索的目標(biāo)檢測(cè)算法。該算法首先對(duì)藥植圖像進(jìn)行高斯濾波去噪,并對(duì)圖像做歸一化預(yù)處理。對(duì)預(yù)處理后的圖片使用基于圖的圖像分割算法進(jìn)行原始分割區(qū)域的劃分,計(jì)算相鄰區(qū)域間的顏色、紋理、大小和交疊相似度。最后根據(jù)相似度進(jìn)行區(qū)域合并,最終得到目標(biāo)區(qū)域。文章圖片數(shù)據(jù)集來自PPBC中國(guó)植物圖像庫(kù)以及作者實(shí)地拍攝。實(shí)驗(yàn)結(jié)果表明,該算法對(duì)有花植物檢測(cè)得分達(dá)到84.3,葉片植物檢測(cè)得分達(dá)到67.86,平均檢測(cè)得分為76.08,較原選擇性搜索算法提升12.64。此外,該算法不需要訓(xùn)練,計(jì)算簡(jiǎn)單,適用性更強(qiáng)。
關(guān)鍵詞: 藥用植物目標(biāo)檢測(cè);選擇性搜索;語義分割;區(qū)域合并
中圖分類號(hào): TP391.41 ? ?文獻(xiàn)標(biāo)識(shí)碼: A ? ?DOI:10.3969/j.issn.1003-6970.2020.06.021
本文著錄格式:孟令濤,孫國(guó)強(qiáng). 基于選擇性搜索的藥用植物目標(biāo)檢測(cè)[J]. 軟件,2020,41(06):96101
【Abstract】: Target detection of medicinal plants can be effectively applied to image recognition and semantic segmentation of medicinal plants. This article optimizes the existing algorithms to make it more accurate for the medicinal plant target detection in the natural environment, and proposes a target detection algorithm based on selective search. The algorithm first performs Gaussian filtering and denoising on the medicine plant image, and performs normalized preprocessing on the image. The pre-processed picture is divided into the original segmentation regions using a graph-based image segmentation algorithm, and the color, texture, size, and overlap similarity between adjacent regions are calculated. Finally, the region is merged according to the similarity, and the target region is finally obtained. The article picture data set is from the PPBC China Plant Image Library and the author field shooting. Experimental results show that the algorithm achieves a score of 84.3 for flowering plants, a leaf plant detection score of 67.86, and an average detection score of 76.08, which is 12.64 higher than the original selective search algorithm. In addition, the algorithm requires no training, is simple to calculate, and more applicable.
【Key words】: Medicinal plant target detection; Selective search; Semantic segmentation; Region merging
0 ?引言
中國(guó)擁有世界上最豐富的藥用植物資源,對(duì)藥用植物的發(fā)現(xiàn)、使用和栽培有著悠久的歷史。我國(guó)地域遼闊,從寒溫帶直到熱帶,地形復(fù)雜,氣候多樣,是世界上植物種類最多的國(guó)家之一,世界已知植物共約27萬種,我國(guó)存在約25700種,其中很多植物具有藥用價(jià)值。20世紀(jì)80年代,我國(guó)曾進(jìn)行過全面系統(tǒng)的資源調(diào)查,發(fā)現(xiàn)我國(guó)的藥用植物資源種類包括383科,2309屬,11146種,其中藻、菌、地衣類低等植物有459種,苔蘚、蕨類、種子植物類高等植物有10687種。……