





摘要 探討了采用預訓練的卷積神經網絡(CNN)模型,如GoogLeNet、VGGNet和EfficientNet,作為特征提取器對蘋果葉片病害檢測準確率的影響。通過結合這3個CNN模型導出的深度特征,實現了深度學習特征的組合,使用提取的深度特征訓練了支持向量機(SVM)分類器。結果表明,所有CNN模型都能以顯著的準確率使用深度特征提取并檢測出蘋果葉片病害,整體分類準確率達到了99.42%。此外,該研究還提出了一種基于改進深度學習的方法,通過結合3個CNN模型的深度特征,進一步提高了預測性能。該方法在蘋果葉片病害檢測中表現出色,并有望應用于其他植物葉片的病害檢測。該研究為植物病害的自動識別提供了一種有效的方法,有助于農業生產的智能化和精準化。
關鍵詞 蘋果葉片病害;卷積神經網絡;深度特征提取;支持向量機;病害檢測
中圖分類號 S-058 文獻標識碼 A 文章編號 0517-6611(2024)23-0216-04
doi:10.3969/j.issn.0517-6611.2024.23.047
Detection of Enhanced Apple Leaf Disease Using Fused Deep Features from Pre-trained CNNs
ZHANG Zheng-feng, GAO Feng
(Xuzhou Vocational College of Bioengineering, Xuzhou, Jiangsu 221006)
Abstract A comprehensive examination of the application of pre-trained Convolutional Neural Networks (CNNs) was discussed, such as GoogLeNet, VGGNet and EfficientNet in detecting apple leaf diseases and pests. By addressing the limitations and gaps in existing research, we focused on enhancing detection accuracy by leveraging deep features extracted from these CNN models. The methodology involved the fusion of deep features obtained from the final fully connected layers of the CNNs, followed by the training of a Support Vector Machine (SVM) classifier. Results showed that all the CNN models demonstrated significant accuracy in detecting apple leaf diseases using deep feature extraction, achieving an overall classification accuracy of 99.42%. Furthermore, an improved deep learning approach was introduced which combined the deep features from the three CNN models, further boosting predictive performance. The methodology exhibited promising results in apple leaf disease detection and had potential applications in detecting diseases in other plant leaves. This research contributed to the development of automated and precise plant disease identification techniques, paving the way for intelligent and targeted agricultural production.
Key words Apple leaf diseases;Convolutional Neural Networks (CNNs);Deep feature extraction;Support Vector Machine (SVM);Disease detection
基金項目 中國高校產學研創新基金項目“基于VR的互聯網-紅色旅游應用研究-以淮海戰役烈士紀念塔園林景區為例”(2022IT061)。
作者簡介 張正風(1980—),男,江蘇徐州人,副教授,碩士,從事軟件工程和軟測量技術研究。*通信作者,副教授,碩士,從事動物繁殖和遺傳育種研究。
收稿日期 2024-04-03
蘋果作為全球公認的種植最廣泛的水果之一,因其營養價值和產量高而受到人們的喜愛,但是蘋果同樣不能免受病害的普遍影響,這對農業產量和質量構成了重大威脅[1]。盡管植物保護專家采用了傳統的視覺檢查方法來檢查植物病害,但這種方法既耗時又往往不準確[2]。……