Numerical Estimation of Mango (Mangifera indica L.) Yield in Côte d'Ivoire Using Integrated Image Analysis and Predictive Modeling
KOFFI Konan Jean-Mathias *
University Jean Lorougnon Guédé, UFR Agroforesterie, Agricultural Production Improvement Laboratory, BP 150 Daloa, Côte d'Ivoire.
YAO Kouadio Jacques-Edouard
University Jean Lorougnon Guédé, UFR Agroforesterie, Agricultural Production Improvement Laboratory, BP 150 Daloa, Côte d'Ivoire.
KANGA Ahou-Nadia
University Jean Lorougnon Guédé, UFR Agroforesterie, Agricultural Production Improvement Laboratory, BP 150 Daloa, Côte d'Ivoire.
AYOLIE Koutoua
University Jean Lorougnon Guédé, UFR Agroforesterie, Agricultural Production Improvement Laboratory, BP 150 Daloa, Côte d'Ivoire.
SORO Dogniméton
University Jean Lorougnon Guédé, UFR Agroforesterie, Agricultural Production Improvement Laboratory, BP 150 Daloa, Côte d'Ivoire.
N’DA ADOPO Achille
National Center for Agricultural Research (CNRA), Regional Directorate of Korhogo, BP 856 Korhogo, Côte d’Ivoire.
*Author to whom correspondence should be addressed.
Abstract
Aims: The present work aims to use a Faster R-CNN network combined with a predictive model to estimate the yield of the mango tree.
Place and Duration of Study: The experiment was conducted in the Poro region of northern Côte d'Ivoire on a mango production campaign. This tropical context, marked by high environmental variability, offers a relevant framework for testing the robustness of the system.
Methodology: This study uses an innovative mixed approach based on image analysis by a convolutional neural network (Faster R-CNN) and a predictive model to automate fruit detection and estimate mango yield. Digital images were captured on two opposite sides of each tree and then analyzed by the neural network. Three linear regression models were developed to correct for biases related to partial fruit visibility.
Results: The results showed an overall efficiency of the network with an F1-score of 0.88 and a detection accuracy of 91 %. The optimal regression model achieved a coefficient of determination (R²) of 0.96 and a normalized error (NRMSE) of 5.9 %. The combination of the network and the corrective model allowed a reliable estimate of the production per tree.
Conclusion: These results demonstrate the potential of artificial intelligence technologies to improve the monitoring of fruit crop yields, and highlight the value of digital solutions adapted to the West African context to strengthen agricultural decision-making and enhance local value chains.
Keywords: Artificial intelligence, CNN, Côte d’Ivoire, image analysis, mango tree, yield