Rice (Oryza sativa L.) is one type of cultivated plant that has the ability to adapt to various conditions and is a main ingredient of food in many countries, especially in Asia, including Indonesia. Although rice can adapt to different environments, it's still susceptible to leaf diseases like brown spot, hispa, and blast. With current technological developments, machine learning is rapidly improving the efficiency of human work in various fields. Computer vision is one of the machine learning methods that is shown to be effective in performing image classification. The application of computer vision, especially through the Convolutional Neural Network (CNN) model, promises to be an effective solution. However, challenges arise in the wild rice plant dataset, where photographs are taken directly from rice fields. To overcome this, data augmentation techniques become important. Data augmentation not only overcomes overfitting but also helps the model understand diverse datasets. This research focuses on using data augmentation methods including Cutmix, Cutout, Mixup, FMix, and All Combination (Cutmix + Cutout + Mixup + FMix) to enhance the performance of the rice leaf detection model. From the test results, the mixed-based data augmentation method significantly improves the performance of the transfer learning model. All Combination method get the maximum accuracy of 91.51% when applied to ResNet34 on Setup Dataset and 99.75% when applied to MobileNetV3 on Wild Dataset. The proposed method can be combined with other methods to create a precise rice leaf disease detection system.

中文翻译：

---

用于检测野生稻叶病的综合混合数据增强


水稻（Oryza sativa L.）是一种能够适应各种条件的栽培植物，是许多国家，特别是包括印度尼西亚在内的亚洲国家的食品主要成分。尽管水稻可以适应不同的环境，但它仍然容易受到褐斑病、稻瘟病和稻瘟病等叶部病害的影响。随着当前技术的发展，机器学习正在迅速提高人类在各个领域的工作效率。计算机视觉是被证明可以有效执行图像分类的机器学习方法之一。计算机视觉的应用，特别是通过卷积神经网络（CNN）模型，有望成为一种有效的解决方案。然而，野生稻植物数据集出现了挑战，该数据集的照片是直接从稻田拍摄的。为了克服这个问题，数据增强技术变得很重要。数据增强不仅可以克服过度拟合，还可以帮助模型理解不同的数据集。本研究重点使用数据增强方法，包括 Cutmix、Cutout、Mixup、FMix 和 All Combination（Cutmix + Cutout + Mixup + FMix）来增强水稻叶片检测模型的性能。从测试结果来看，基于混合的数据增强方法显着提高了迁移学习模型的性能。所有组合方法在Setup数据集上应用于ResNet34时获得了91.51%的最大准确率，在Wild数据集上应用于MobileNetV3时获得了99.75%的最大准确率。该方法可以与其他方法结合创建精确的水稻叶部病害检测系统。