Objects detection on images using machine learning
Loading...
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Данная работа посвящена решению задачи детектирования сигарет на изображениях. Для этого обучена модель YOLOv5 на собранных и размеченных вручную 1500 изображений, содержащих сигареты. В ходе работы исследована возможность повышения точности детектирования за счет аугментации данных и модификации функции потерь. Также предложен дополнительный модуль классификации для уменьшения ошибок второго рода. Классификатор обучен на 1400 изображениях сигарет и объектов, похожих на сигареты. Проведенные эксперименты показывают увеличение точности Mean Average Precision.
This work is devoted to solving the problem of detecting cigarettes in images. For this, the YOLOv5 model was trained on 1500 manually collected and labeled images containing cigarettes. In the course of the work, the possibility of increasing the accuracy of detection by augmenting the data and modifying the loss function was studied. An additional classification module is also proposed to reduce type II errors. The classifier is trained on 1400 images of cigarettes and cigarette-like objects. The conducted experiments show an increase in Mean Average Precision.
This work is devoted to solving the problem of detecting cigarettes in images. For this, the YOLOv5 model was trained on 1500 manually collected and labeled images containing cigarettes. In the course of the work, the possibility of increasing the accuracy of detection by augmenting the data and modifying the loss function was studied. An additional classification module is also proposed to reduce type II errors. The classifier is trained on 1400 images of cigarettes and cigarette-like objects. The conducted experiments show an increase in Mean Average Precision.