Improving the effectiveness of training the on-board object detection system for a compact unmanned aerial vehicle
Journal Title: Восточно-Европейский журнал передовых технологий - Year 2018, Vol 4, Issue 9
Abstract
<p class="a"><span lang="EN-US">The model of object detector and the criterion of leaning effectiveness of the model were proposed. The model contains 7 first modules of the convolutional Squeezenet network, two convolutional multiscale layers and the informationextreme classifier. The multiplicative convolution of the particular criteria that takes into account the effectiveness of detection of objects in the image and accuracy of the classification analysis was considered as the criterion of learning effectiveness of the model. In this case, additional use of the orthogonal matching pursuit algorithm in calculating highlevel features makes it possible to increase the accuracy of the model by 4 %. The training algorithm of object detector under conditions of a small size of labeled training datasets and limited computing resources available on board of a compact unmanned aerial vehicle was developed. The essence of the algorithm is to adapt the highlevel layers of the model to the domain application area, based on the algorithms of growing sparse coding neural gas and simulated annealing. Unsupervised learning of highlevel layers makes it possible to use effectively the unlabeled datasets from the domain area and determine the required number of neurons. It is shown that in the absence of fine tuning of convolutional layers, 69 % detection of objects in the images of the test dataset Inria Aerial Image was ensured. In this case, after fine tuning based on the simulated annealing algorithm, 95 % detection of the objects in test images is ensured. </span></p><p class="a"><span lang="EN-US">It was shown that the use of unsupervised pretraining makes it possible to increase the generalizing ability of decision rules and to accelerate the iteration process of finding the global maximum during supervised learning on the dataset of limited size. In this case, the overfitting effect is eliminated by optimal selection of the value of hyperparameter, characterizing the measure of coverage of the input data of by network neurons. </span></p>
Authors and Affiliations
Vyacheslav Moskalenko, Anatoliy Dovbysh, Igor Naumenko, Alyona Moskalenko, Artem Korobov
Keywords
Related Articles
- EP ID EP528108
- DOI 10.15587/1729-4061.2018.139923
- Views 62
- Downloads 0
Development of the mathematical model and the method to solve a problem on the optimization of packing the ellipsoids into a convex container
<p>This paper addresses the problem on the optimal packing of the predefined set of ellipsoids into a convex container of minimum volume. The ellipsoids are assigned by the dimensions of semi-axes and arrangement paramet...
Analysis of the interaction between properties and microstructure of construction ceramics
<p>We have defined the basic properties of ceramic bricks for various purposes. The microstructure and phase composition of ordinary, facing and clinker bricks were investigated. We have established the interrelation bet...
Increasing efficiency of plasma hardening by local cooling of surface by air with negative temperature
<p>The martensitic transformation interval of some hypoeutectoid, all eutectoid and all hypereutectoid steels covers to a large extent the region of negative temperatures. Due to the fact that the plasma hardening operat...
Optimizing performance of lithiumion battery by nanosilicon addition mixed in LI4TI5O12 anode made using mechanochemicalhydrothermal method
<p>Lithium Titanate (Li<sub>4</sub>Ti<sub>5</sub>O<sub>12</sub> or LTO) is one of the best candidates to replace graphite as anode material in the lithium-ion battery (LIB), due to unwanted solid electrolyte interphase (...
Studying the physicalchemical transformations at resourcesaving reduction melting of chrome–nickelcontaining metallurgical waste
<p>We have investigated the physicochemical characteristics of chromium-nickel-containing waste from the production of corrosion-resistant steels and a doped alloy obtained by reduction smelting. This is necessary to det...