Computer Vision Approaches For Smart Farming: Yield Prediction and Monitoring

Abstract

Smart farming is a combination of sophisticated technologies that are used to maximize agricultural output and sustainability. Computer vision is one of these technologies that have become a force to be used in automating crop monitoring and yield prediction. This paper discusses the different computer vision techniques applied in agricultural yield forecasting, and primarily looks at deep learning-based techniques applied to both aerial and ground images. The research focuses on what the recent literature indicates as the effectiveness of the convolutional neural networks (CNNs), vision transformers (ViTs), and object detectors as a form of estimation of yield in large crops through the analysis of secondary data and model performance. The methodology entails a comparative study of the techniques of counting the images of the plant, estimating the canopy cover as well as the technique of biomass modeling. It has been shown that CNN-based methods have an average accuracy in yield predictions of 88 percent and ViT models outperform by up to 93 percent on the same metric as they are able to capture long-range correlations in aerial images. UAV imagery used along with computer vision methods will greatly minimize the time spent in data collection using manual means and enhance farm management decisions. The paper concludes that precision agriculture can be assisted by the adoption of computer vision approaches since it facilitates early, scalable, and accurate yield prediction, which eventually will improve productivity and sustainability.