Tea leaf disease recognition using attention convolutional neural network and handcrafted features

Abstract

The diseases of tea leaves have a significant impact on their quality and yield, making the rapid identification of leaf diseases in tea crucial for prevention and control. We propose an LBPAttNet model, incorporating a lightweight coordinate attention mechanism into ResNet18 to enhance disease localization and reduce background interference. Furthermore, we employ the local binary patterns (LBP) algorithm to further extract local structural and textural features of tea leaf diseases, and integrate deep features to obtain a more comprehensive feature representation. Additionally, we utilize the focal loss function to alleviate the issues of class imbalance and varying difficulty levels in tea leaf disease, thereby further enhancing the accuracy of tea disease recognition. Our model achieves an accuracy of 92.78% and 98.13% on two publicly available tea disease datasets, surpassing ResNet18 by 3.84% and 2.59% respectively. Compared to traditional algorithms such as AlexNet, GoogleNet, MobileNet, VGG16, and other tea disease recognition algorithms, our model also shows significant improvements. These results highlight the superior performance and robustness of our model.

Plant phenotyping relevance

茶葉の病徴を画像から認識するCNNと特徴抽出手法を開発し、公開データセットで性能評価しているため、植物病害状態のフェノタイピング手法が中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.