Aplicación de Machine Learning para estimar la producción de empresas agrícolas exportadoras : Modelo de Machine Learning para identificar los 2 primeros estados fonológicos de un producto agrícola a exportarse.

Abstract

The agricultural export sector is one of the sectors where technology has not been implemented or few people have sufficient motivation to enter this field, for this reason there are problems that require a technological and scientific solution. One of the problems is that it is essential for the agricultural export sector to know the production it has on its land. Currently this control is carried out operationally by industry workers, which has a high margin of error since many of the roses measure around 3 meters, which is a problem for humans to detect the entire production. The document presents a possible solution to correct counting problems and improve the times it takes to perform this task using Machine Learning and Deep Learning techniques to create an AI model that performs this task. To develop the AI model, pre-trained neural network models such as Yolov5, Yolov6, Yolov8, among others, were used. These models are already pre-trained with large data datasets that provide a base knowledge to classify, detect and segment objects of our interest, in our case, roses. However, not everything is solved since it is necessary to know the number of roses depending on their phonological state and for this a technique called ByteTrack was used that provides us with tools that facilitate the counting and tracking of a rose during its movement in the video provided.