Automated Grapevine Cultivar Identification via Leaf Imaging and Deep Convolutional Neural Networks: A...

Area of Science:

• Agricultural Science
• Computer Science
• Plant Biology

Background:

• Grapevine (Vitis vinifera) cultivation involves thousands of cultivars, traditionally identified by expert ampelography.
• Current identification methods like molecular genetics can be costly and data-limited.
• Accurate and efficient cultivar identification is crucial for viticulture and grapevine breeding.

Purpose of the Study:

• To develop an automated system for grapevine cultivar identification using convolutional neural networks (CNNs).
• To leverage leaf images in the visible spectrum for distinguishing between numerous grapevine varieties.
• To provide a rapid, low-cost, and high-throughput solution for grapevine identification.

Main Methods:

• A modified VGG16 CNN architecture was employed, incorporating global average pooling, dense layers, batch normalization, and dropout.
• The model was trained and evaluated using leaf images captured in the visible light spectrum (400-700 nm).
• A five-fold cross-validation strategy was utilized to assess model performance and predictive efficiency.

Main Results:

• The developed CNN model accurately distinguished between diverse grapevine varieties based on intricate visual leaf features.
• The model achieved an average classification accuracy exceeding 99% across different grapevine cultivars.
• The system demonstrated high predictive efficiency and low uncertainty in cultivar identification.

Conclusions:

• The automated CNN-based system offers a significant advancement for grapevine cultivar identification.
• This tool complements traditional ampelography and quantitative genetics, enhancing cultivar identification services.
• The proposed method provides a rapid, cost-effective, and scalable approach for the viticulture industry.