Using AI Technology to Detect Black Rot on Grape Leaves

Jim Crocker
16th April, 2025

Using AI Technology to Detect Black Rot on Grape Leaves

The dataset used to train and test the black rot detection model included images of infected grape leaves from the controlled Plant Village collection (A) and a real-world orchard environment (B).

Image adapted from: Zhu et al. / CC BY (Source)

Key Findings

  • In Nantong, China, researchers created an advanced AI tool that accurately detects grape leaf diseases
  • The improved system can identify small disease spots and works well with low-quality images
  • This new method outperforms previous models, helping farmers manage grape diseases more effectively
Grape cultivation faces significant challenges from various pests and diseases, which can severely impact agricultural productivity and economic stability. Timely and accurate identification of these diseases is crucial for effective management and prevention of widespread damage. Recent advancements in artificial intelligence, particularly in deep learning, have shown promise in enhancing disease detection in crops. A study conducted by researchers at Nantong Institute of Technology, Nantong, China[1], introduces an improved recognition network based on YOLO v8 to address common issues in grape leaf disease detection. Traditional methods for monitoring grape leaf diseases often rely on manual inspection, which is not only time-consuming and labor-intensive but also prone to human error. Previous research has demonstrated the effectiveness of convolutional neural networks (CNNs) in diagnosing grape leaf diseases with high accuracy[2][3][4]. For instance, models like DICNN and ECA-SNet have achieved impressive accuracy rates by leveraging large datasets and sophisticated network architectures. However, challenges remain, particularly in accurately detecting small disease spots and maintaining high detection accuracy with low-resolution images. The study from Nantong Institute of Technology builds on these earlier findings by enhancing the YOLO v8 model, a popular object detection framework known for its speed and accuracy. The researchers identified that YOLO v8 struggled with recognizing small target spots and suffered from low detection accuracy when dealing with low-resolution input images. To overcome these limitations, they introduced two key modifications to the network. Firstly, the traditional convolution and pooling layers in YOLO v8 were replaced with Spatial Pyramid Dilated Convolution (SPD-Conv). This advanced convolution technique allows the network to capture detailed features of small targets more effectively by expanding the receptive field without losing resolution. The SPD-Conv enhances the network's ability to recognize fine-grained details, which is essential for identifying early-stage disease symptoms that manifest as small spots on grape leaves. Secondly, the researchers incorporated an Efficient Multi-Scale Attention (EMA) Module into the Neck part of the YOLO v8 architecture. The EMA module optimizes the use of feature information across different detection layers, improving the overall accuracy of feature representation. This enhancement ensures that the model can better distinguish between various disease patterns by focusing on the most relevant parts of the image data. To evaluate the performance of the improved YOLO v8 model, the study utilized the Plant Village dataset along with a set of orchard images. The Plant Village dataset is a comprehensive collection of plant images used extensively in agricultural research for training and testing disease detection models. The inclusion of real-world orchard images ensured that the model's performance was assessed in practical scenarios. The experimental results were promising. The improved YOLO v8 model achieved a precision of 92.64%, a recall of 93.28%, and an Average Precision (AP) of 96.17%. Additionally, the model size was significantly reduced to just 7.1 million parameters, making it lightweight and suitable for deployment in real-world agricultural settings where computing resources may be limited. Compared to the original YOLO v8, the improvements in precision, recall, and AP were 2.38%, 1.91%, and 1.13%, respectively. When benchmarked against other mainstream networks such as YOLO v4, YOLO v5, YOLO v6, and YOLO v7, the enhanced YOLO v8 demonstrated superior precision improvements of 4.74%, 3.38%, 4.15%, and 4.69%, respectively. These enhancements not only improve the detection accuracy of small target objects but also enable more precise identification of specific diseases like black rot, which is particularly detrimental to grape leaves. Black rot causes significant economic losses by reducing grape yields and quality. Accurate detection allows for timely intervention, minimizing the spread and impact of the disease. This study aligns with and extends the findings of previous research that utilized CNN-based models for grape disease detection. For example, the DICNN model achieved high accuracy by employing an Inception structure and dense connectivity strategy[2], while ECA-SNet focused on creating a lightweight model with efficient attention mechanisms[3]. The improved YOLO v8 model integrates similar principles of enhancing feature extraction and attention mechanisms but applies them within the context of an object detection framework tailored for real-time monitoring. Furthermore, the DCNN Classifier model based on VGG16, which demonstrated high accuracy and the potential to serve as a decision support system for farmers[4], shares the common goal of leveraging deep learning for practical agricultural applications. The improved YOLO v8 model complements these approaches by offering a robust tool for real-time dynamic monitoring, which is essential for maintaining the health and productivity of grape orchards. In conclusion, the research from Nantong Institute of Technology presents a significant advancement in the field of agricultural technology by improving the YOLO v8 network for grape leaf disease detection. The integration of Spatial Pyramid Dilated Convolution and Efficient Multi-Scale Attention modules addresses key challenges in recognizing small disease spots and maintaining high accuracy with low-resolution images. This enhanced model not only builds on the successes of previous CNN-based approaches but also provides a practical solution for real-time disease monitoring, ultimately contributing to the sustainable development of the grape industry.

AgriculturePlant Science

References

Main Study

1) An application of YOLOv8 integrated with attention mechanisms for detection of grape leaf black rot spots

Published 15th April, 2025

https://doi.org/10.1371/journal.pone.0321788

Related Studies

2) Grape Leaf Disease Identification Using Improved Deep Convolutional Neural Networks.

https://doi.org/10.3389/fpls.2020.01082

3) Fine-Grained Grape Leaf Diseases Recognition Method Based on Improved Lightweight Attention Network.

https://doi.org/10.3389/fpls.2021.738042

4) Multiclass classification of diseased grape leaf identification using deep convolutional neural network(DCNN) classifier.

https://doi.org/10.1038/s41598-024-59562-x


Related Articles

An unhandled error has occurred. Reload 🗙