Literary review 1 while writing on neural networks and drones

While writing a paper on the uses of neural networks and drones I learned a lot about the advantages and disadvantages of them. These allow for a better understanding of when it is appropriate to use one with a drone and when it is ineffective to use one.

Advantages
The main advantages with using a neural network come from using a vision based system to aid the drone in operation. The vision system can be used for tracking people, perfecting landings, and even counting cows. The benefits of using a neural network is that it can reduce the complexity of operations, or it can improve the efficiency of the operations. These are the main functions for any additional software for drones, and a neural network is a much more specific software. In applications like counting cows, the neural network is ground based, and is just processing on a computer to analyze the data post flight. It does this through image filtering, figure 1, and that allows the computer to count the number of cows. The neural network makes the process faster than just a human going out and counting them is that the terrain in the Spanish mountain side makes the process of checking a field take much longer on foot than with a drone.

Figure 1. The post processing from a raw photo (top left) to
one that the neural network can use to count (bottom left)

Disadvantages
I learned that there are a bunch of disadvantages to neural networks that make them hard to use with suas. These disadvantages can be about the hardware needed to run a neural network or coding the neural network. For the hardware, neural networks need a lot of ram to run, this increases the current draw if they are onboard the aircraft, or it increases the cost of operation if they are on the ground system. Another disadvantage is that coding an effective neural network takes lots of time and with a vision system it takes lots of images from different settings to effectively be able to constantly collect data. 

Discussion
I think that the neural networks could play a much more important role in the contingency plans for a uav as they could potentially decide where to land in the case of a failure. They could also learn more about the aircraft as more flight hours happen and predict failures before they happen

Conclusion
 In conclusion I think that we will see a lot more uavs with neural network powered features as the processing capabilities increase, and I think we will also see more neural-networks doing post processing on the ground before we see them on the uavs.