Intelligent yield estimation for tomato crop using SegNet with VGG19 architecture.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate crop yield estimation (YE) is crucial for effective fruit management and marketing.
• Manual YE methods are time-consuming, subjective, and require expert knowledge, posing limitations for large-scale agriculture.

Purpose of the Study:

• To develop an intelligent system for automated tomato yield estimation.
• To overcome the limitations of manual crop YE through deep learning-based image analysis.

Main Methods:

• A SegNet with VGG19 deep learning architecture was employed for semantic segmentation to detect, localize, and count tomatoes.
• The model was trained on a dataset of 672 images and compared with U-Net and SegNet with VGG16 architectures.
• A case study was conducted in a real tomato field for model validation.

Main Results:

• The proposed SegNet with VGG19 model demonstrated superior performance compared to other semantic segmentation architectures.
• The system achieved test precision of 89.7%, recall of 72.55%, and F1-score of 80.22%.
• The method showed reasonable localization accuracy for tomatoes in field conditions.

Conclusions:

• The developed intelligent YE system offers a more efficient and objective approach to tomato yield estimation.
• Deep learning-based semantic segmentation holds significant potential for enhancing agricultural management practices.