WP3-36 2

2022-03-13T06:55:30Z

Udanet: Created page with "=AI drone system modules= {|class="wikitable" | ID|| WP3-03 |- | Contributor || UDANET |- | Levels || Functional |- | Require || Dataset |- | Provide || Designed and implemented algorithm |- | Input || Plant images |- | Output || Disease classification |- | C4D building block || Data analytics |- | TRL || 4-5 |} ==Detailed Description== An artificial intelligent method to classify leaf diseases will be designed and implemented. The methods inputs a..."

=AI drone system modules=
{|class="wikitable"
| ID|| WP3-03
|-
| Contributor || UDANET
|-
| Levels || Functional
|-
| Require || Dataset
|-
| Provide || Designed and implemented algorithm
|-
| Input || Plant images
|-
| Output || Disease classification
|-
| C4D building block || Data analytics
|-
| TRL || 4-5
|}

==Detailed Description==
An artificial intelligent method to classify leaf diseases will be designed and implemented. The methods inputs are images from cameras, and the output is the plant health status classification. Artificial intelligence algorithms with different characteristics will be designed and implemented with the aim of taking into account the computational cost of each, as well as the over-fitting problem and the dataset that is not always adequate, and the diagnostic performance is drastically decreased when used on test datasets from new environments (i.e. generalization problem).

A large amount of data increases the performance of machine learning algorithms and avoids over-fitting problems. Creating a large amount of data in the agricultural sector for the design of models for the diagnosis and detection of plant diseases is an open and challenging task that takes a lot of time and resources. One way to increase the dataset will be to use data augmentation techniques. It increases the diversity of training data for machine learning algorithms without collecting new data.

==Contribution and Improvements==
The improving objective of this component is to develop and test the algorithms, increase the reference dataset using basic image manipulation and deep learning based image augmentation techniques such as image flipping, cropping, rotation, colour transformation, PCA colour augmentation, noise injection, Generative Adversarial Networks (GANs) and Neural Style Transfer (NST) techniques. Performance of the data augmentation techniques was studied using state of the art transfer learning techniques both in terms of accuracy and computational cost.

==Design and Implementation==
It is developed through the Tensorflow and Python framework.

WP3-36 1

2022-03-13T06:52:54Z

Udanet: Created page with "=Smart and predictive energy management system= {|class="wikitable" | ID|| WP3-03 |- | Contributor || UDANET |- | Levels || Functional |- | Require || Sensors that provide measurements and mission details |- | Provide || Energetic trajectories to perform path tracking missions |- | Input || X, Y, Z coordinates of the drone and its velocities Roll, pitch, and yaw angles of the drone Initial and final position of the mission |- | Output || Rotor’s speed (..."

=Smart and predictive energy management system=
{|class="wikitable"
| ID|| WP3-03
|-
| Contributor || UDANET
|-
| Levels || Functional
|-
| Require || Sensors that provide measurements and mission details
|-
| Provide || Energetic trajectories to perform path tracking missions
|-
| Input || X, Y, Z coordinates of the drone and its velocities
Roll, pitch, and yaw angles of the drone
Initial and final position of the mission
|-
| Output || Rotor’s speed (or forces)
|-
| C4D building block || Flight planning
|-
| TRL || 3-4
|}

==Detailed Description==
An energy management system is vital to optimize the energy life and the purpose of the system. It will continuously monitor important system parameters, while dealing with the varying power demands of the many aspects, the objectives of the mission and optimizing the usage of the energy. Given the initial and final positions, the objective of the component is to compute the control inputs that rule the motion and vehicle trajectory to optimize energy consumption and the computational burden of the algorithm.

The aim of this activity is to use a model-based approach to control UAV flight to minimize energy consumption. It is therefore necessary to consider the flight dynamics, battery, and energy flow and actuator models. The strategy that led to good results in some research is the optimal control and therefore this type is considered in the activity. A first objective is to improve control by making it robust with respect to model approximation errors or disturbance (for example under wind conditions).

==Contribution and Improvements==
The main contribution of the component is to provide excellent trajectories from an energy point of view for the position, speeds that are the result of an elaboration of the associated optimal control problem and analysis of the excellent results. The improvement is due to the fact that a real-time resolution of the control is not required, but rule-based strategies are required which therefore have a low computational cost. Furthermore, in collaboration with other partners, they want to experiment in non-ideal conditions (presence of wind)

==Design and Implementation==
The designed energy management system will be verified and tested via Software in The Loop using software Matlab-Simulink. The reference generator will be implemented to execute a mission and the associated controller and will be tested both with Matlab-Simulink and with complex simulators in less than ideal conditions.

COMP4DRONES - User contributions [en]

WP3-36 2

WP3-36 1