A Multiple Instance Dictionary Learning Approach for Corn Yield Prediction From Remote Sensing Data

Corn occupies a significant portion of American residents’ diet. Therefore, accurate prediction of corn’s annual yield in agricultural cultivation would greatly assist farmers in improving planting efficiency and has significant implications for agricultural resource management, market planning, food safety monitoring, and other related fields. To enhance the accuracy of corn yield prediction, this research utilized remote sensing satellite data and employed the multiple instance online dictionary learning (MIDL) method to predict county-level corn yield within 12 states in the Midwest region of U.S. MIDL combines multiple instance learning to retain detailed information within each county and dictionary learning (DL) to filter and eliminate potentially interfering mixed pixels’ information in the prediction process. Experimental results demonstrate that MIDL achieved high prediction accuracy and exhibited excellent spatial generalization capabilities, outperforming all the compared methods. This study extensively analyzed the strengths and weaknesses of MIDL, confirming its promising potential, and identified directions for future improvements. Proposed MIDL method had the best performance for the six testing years 2016–2021, achieving an average <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$ \text {R}^{2} $ </tex-math></inline-formula> of 0.79.