Determinación de la madurez de frutos del café mediante el reconocimiento de imágenes utilizando un modelo basado en redes neuronales convolucionales

The province of Loja is characterized by producing high altitude coffee (Arabica species) which involves good agronomic management throughout the process, this case study focuses on the harvest phase and the conventional techniques used to determine the optimal maturity of the fruit which tend to be subjective and even the tools used tend to be expensive, therefore it is intended to help these traditional methods to make an accurate identification, which contributes to enhance the quality of cup, through artificial vision, providing a non-destructive method with a high level of accuracy. For this reason, the present work proposes the creation of an object detection model based on the architecture of Convolutional Neural Networks, to identify coffee fruits in their four stages of maturation: immature, pinto, ripe and overripe. In order to obtain the data set for the training, validation and testing phases, field work was carried out where a series of photographs containing the ripening cycle of the Colombia 6 variety were captured, and the data augmentation technique called "image rotation" was applied to them, with which a final dataset of 800 images was obtained; these were manually labeled with the LabelImg tool. In the construction phase the Resnet-50, Inception-v2 and Yolov4 models were implemented, from which a mean average precision value(mAP) of 64.85%, 67.37% and 97.59% respectively was obtained, thus defining that the model with a superior performance is Yolov4 with the Darknet architecture, due to this the model was trained for a longer period of time obtaining a precision index of 93%, sensitivity of 96%, f1-score of 95% and mAP of 95. 01% (object detection metrics) in addition to obtaining confusion matrix metrics with a level of precision, accuracy, sensitivity and specificity greater than 95%, surpassing the results presented in related works. Keywords: Object detection, Coffee fruits, Inception-v2, Resnet-50, Yolov4.

• প্রধান লেখক: Jiménez Cueva, Karla Yacely (author)
• বিন্যাস: bachelorThesis
• ভাষা: spa
• প্রকাশিত: 2023
• বিষয়গুলি: <DETECCIÓN DE OBJETOS> < FRUTOS DE CAFÉ> < INCEPTION-V2> < RESNET-50> < YOLOV4>
• অনলাইন ব্যবহার করুন: https://dspace.unl.edu.ec/jspui/handle/123456789/26414