萬路瑤 葉安勝



摘 要:為促進農業向精準農業、智慧農業方向發展,針對種業中衡量種子質量的重要指標千粒重,為提高其精確度,提出基于非負矩陣分解與支持向量機的粘連種子分類算法,在利用機器視覺與圖像識別等技術進行種子顆粒計數時,解決種子因嚴重粘連(經圖像預處理、形態學操作等仍粘連)使系統出現誤判,導致計數結果誤差較大的問題。實驗以玉米種子為研究對象,研究3種常見粘連類型。采用基于NMF的方法實現種子粘連圖像有效特征的提取,再運用SVM構建多分類器,解決種子粘連類型的三分類問題,最后通過實驗驗證,該方法的分類準確率為98%。
關鍵詞:非負矩陣分解;支持向量機;種子粘連;種子分類
DOI:10. 11907/rjdk. 191464 開放科學(資源服務)標識碼(OSID):
中圖分類號:TP317.4 文獻標識碼:A 文章編號:1672-7800(2020)002-0153-04
英標:Classification of Adhesive Seeds Based on NMF and SVM
英作:WAN Lu-yao, YE An-sheng
英單:(School of Information Science & Engineering, Chengdu University, Chengdu 610106, China)
Abstract: In order to respond to the national call to develop agriculture in a precise and smart way, and to measure the importance of seed quality in the seed industry and improve its accuracy, this paper proposes a classification of adhesion seeds based on non-negative matrix factorization and support vector machine. The algorithm solves the problem that the seeds are seriously adhered (image pre-processing, morphological operation, etc. still adheres) when using the techniques of machine vision and image recognition to count the seed particles, so that the error of the counting result caused by the misjudgment of the system is large. The experiment used corn seeds as the research object to study three common types of adhesions. The NMF-based method is used to extract the effective features of seed adhesion images, and then use SVM to construct multi-classifiers to solve the three-class problem of seed adhesion types. The accuracy of the classification was verified to be 98% by experimenting with 90 images.
Key Words: non-negative matrix factorization; support vector machine; seed adhesion; seed classification
0 引言
如今鄉村振興已上升至國家戰略,隨著信息技術的發展,農業信息化也加速推進,農業開始朝著精準農業與高效農業方向發展。種業是農業發展的決定性因素,育種創新是農業轉變發展方式的前提。種子質量的提升與育種技術的發展,有利于保障國家糧食安全,促進農業增效、農民增收等。種子千粒重與種子質量有著直接關系,在千粒重指標獲取與育種工作中,種子的準確計數十分關鍵[1]。
傳統人工計數方法效率與精確度低,之后出現了光電管計數方法如宋礽蘇等[2]設計的自動光電數粒儀等,但若種子在通過光電管時存在部分交疊,很可能產生少計、漏計等情況,且成本高、過程繁瑣?!?br>