Aid System for Estimating Agricultural Yield Using a Deep Learning Technique: Tomato Case

Abstract

The precision of traditional methods for estimating crop yield is a major challenge, particularly for large areas. To improve this process, we developed a tomato detection and localization system using deep learning techniques. The system uses Faster-RCNN, a cutting edge technology of object detection model, to detect and localize tomatoes in images. We trained the model on a database of 150 images, which were normalized to 100*100 pixels in RGB. The system estimates the real sizes of tomatoes using the Ground Sampling Distance method and predicts their masses using a regression model. The model produces an average absolute error of 42.365% and a quadratic error of 51.044%. Our system provides a more efficient and accurate way to estimate tomato crop yields on a large scale.