Deep Learning Regression Approaches Applied to Estimate Tillering in Tropical Forages Using Mobile Phone Images.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate estimation of forage regrowth density is crucial for sustainable livestock management.
• Traditional methods for assessing forage regrowth can be labor-intensive and time-consuming.
• Mobile phone imaging offers a potentially accessible and cost-effective data collection method.

Purpose of the Study:

• To evaluate the performance of Convolutional Neural Network (CNN)-based approaches for estimating forage regrowth density.
• To determine the feasibility of using mobile phone images for this estimation task.
• To compare the effectiveness of different CNN architectures.

Main Methods:

• A dataset of 1124 labeled mobile phone images was created, taken 7 days post-harvest.
• Six CNN architectures were evaluated: AlexNet, ResNet (18, 34, 50 layers), ResNeXt101, and DarkNet.
• The models were trained and validated to predict regrowth density using regression analysis.

Main Results:

• The best performing regression model achieved a mean absolute error of 7.70.
• A high correlation coefficient of 0.89 was observed between predicted and actual regrowth density.
• Several CNN architectures demonstrated significant potential for this application.

Conclusions:

• Deep learning models, particularly CNNs, are effective for estimating forage regrowth density from mobile phone images.
• This approach provides a viable, technology-driven solution for precision agriculture in tropical forage management.
• The use of mobile phones democratizes access to advanced forage assessment tools.