Estimating Optimal Harvest Time and Yield in Tomatoes Using Deep Learning Techniques: A Preliminary Study

The precise estimation of harvest timing and production in greenhouse crops is crucial for farmers and agricultural cooperatives, especially in a context marked by environmental challenges such as climate change, scarcity of natural resources, and economic challenges. Traditionally, this estimation has been done using conventional methods that rely on observation and visual experience, which can lead to inaccurate and subjective estimations. Artificial intelligence (AI) and Deep Learning (DL) models can play a fundamental role as they can identify complex patterns and relationships that help predict the optimal harvest timing and production more accurately. The objective of this work has been to implement a DL model based on CNNs (Convolutional Neural Networks) and create an LSTM (Long Short-Term Memory) model so that, using tomato plant images and data on the number and weight of tomatoes, the optimal harvest timing and production could be estimated. In the case of production estimation, the mean absolute error (MAE) obtained with DL was 0.151 kilograms compared to the actual production. For the estimation of the optimal harvest timing, the error generated by the DL model is set, in some cases, at one day relative to the actual harvest day in the field. These results demonstrate the great potential of the methodology for estimating the optimal harvest timing and production in greenhouse tomato plants.