Aims and background

Pests and diseases of plants often threaten the availability and safety of plants for human consumption. To face these challenges, a new agricultural revolution is underway (agriculture 4.0). This agrarian revolution dramatically benefits from new digital technologies and artificial intelligence (AI).

Methods

The farmers need a reliable tool for an early disease diagnosis. Imaging is a promising technique for diagnosing and quantifying the disease plot. Easily automated and non-intrusive, imaging allows, with low costs in instrumentation and human resources, to account for much agricultural priority’s local mics on large production areas. The main purpose paper is to develop a hybrid model for tomato disease detection based on image data collection. We apply transfer learning and fine-tuning strategies to improve the performance of different pre-trained models. Two models have been selected to develop our hybrid model for plant disease identification among these CNN models. We used the plant village dataset, which contains nine classes of tomato diseases.

Results

First, we evaluate the performance of seven different architectures including VGG16, ResNet50, EfficientNetB0, EfficientNetB1, EfficientNetB2, EfficientNetB3 and EfficientNetB4. We applied the transfer learning technique. Then, the best two pre-trained models were selected and used to implement a weighted average ensemble. The proposed model achieves an accuracy of 0.981.

Conclusion

Many diseases can affect tomato plants and cause yield losses. Therefore, plant pathogens should be given more importance. Furthermore, this study can be adapted to cover other types of crops in future research.

中文翻译：

---

基于遗传算法和深度神经网络的番茄叶片病害检测高效技术设计

目标和背景

植物病虫害经常威胁人类食用植物的可用性和安全性。为了应对这些挑战，一场新的农业革命正在进行中（农业 4.0）。这场农业革命极大地受益于新的数字技术和人工智能 (AI)。

方法

农民需要一种可靠的工具来进行疾病的早期诊断。成像是一种很有前途的技术，用于诊断和量化疾病图。易于自动化和非侵入性的成像允许以低成本的仪器和人力资源，在大型生产区占大部分农业优先的本地麦克风。主要目的是开发一种基于图像数据收集的番茄病害检测混合模型。我们应用迁移学习和微调策略来提高不同预训练模型的性能。选择了两个模型来开发我们在这些 CNN 模型中进行植物病害识别的混合模型。我们使用了植物村数据集，其中包含九类番茄病害。

结果

首先，我们评估了七种不同架构的性能，包括 VGG16、ResNet50、EfficientNetB0、EfficientNetB1、EfficientNetB2、EfficientNetB3 和 EfficientNetB4。我们应用了迁移学习技术。然后，选择最好的两个预训练模型并用于实现加权平均集成。所提出的模型达到了 0.981 的准确度。

结论

许多疾病会影响番茄植株并导致产量损失。因此，应更加重视植物病原体。此外，这项研究可以适用于在未来的研究中涵盖其他类型的作物。