UAV Payload Detection Using Deep Learning and Data Augmentation

Ku, Ilmun; Roh, Seungyeon; Kim, Gyeongyeong; Taylor, C. James; Wang, Yaqin; Matson, Eric T.

doi:10.1109/irc55401.2022.00009

Cited by 6 publications

(1 citation statement)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, Cross-modality evaluation was implemented for infrared images with a performance of 82.8% mAP. Classification of unloaded and loaded with multiple objects drones based on an audio data set by Deep Learning methods was presented in [25]. The data set was the sound realized through flight time while having payloads.…”

Section: Introductionmentioning

confidence: 99%

Payload Prediction for Quadro Drone using Temporal Deep Machine Learning Models

Zitar,

Kashkash,

Seghrouchni

et al. 2024

Preprint

View full text Add to dashboard Cite

In this paper, four different deep-learning and temporal machine-learning techniques are used to predict the payload of the DJI Matrice 100 quadcopter drone based on its tracking data. Tracking variables for real-life experimentations are provided as open source for payloads of 0.0, 250, 500, and 750 grams. The drone is a Quado drone DJI matrice 100. The Machine Learning techniques are RNN, LSTM, TCN, and GRU. The values of the tracks’ kinematics come from several different flights for the different loads. The tracks’ kinematics and flight parameters are used in training the four models. The temporal nature of the data values triggers the need for machine learning methods that use history/memory as part of inputs. This application is a real and relevant test to evaluate the capabilities of the 4 famous types of deep temporal machine learning models. The original data has more than 1400 records for every flight with more than 22 variable values. More than 270 flights were conducted with the 4 payloads. The models were tuned and initialized with the optimum parameters found. Several simulations were implemented for training, testing, and validation. Cross-validation for the models was implemented The RMSE and MAE errors were used to evaluate the loss and the accuracies. The TCN model with the optimum parameters showed the best performance compared to other techniques. The RNN (many-to-one) showed the lowest performance. The results emphasize the superiority of the modern TCN methods over other temporal deep techniques for this kind of problem.

show abstract

Section: Introductionmentioning

confidence: 99%