Development Of Real-time Bird Repellent System Based On Deep Learning

Birds are strongly associated with agricultural production activities and play an important role in exterminating pests of agricultural,but they also cause heavy loss in agricultural economy such as pecking seed crops.Every year,the problem of agricultural bird damage has caused huge economic losses to agricultural production,and bird invasion of farmland will bring serious impact on agricultural food safety.So it is important for the development of agricultural economy to prevent Agricultural bird damage.At present,the ways to prevent birds damage are mainly divided into passive birds-preventing and active birds-driving,the passive birds-preventing measures include: laying anti-bird mesh,planting bait crops,bagging and so on.These methods have the disadvantages of requiring regular maintenance,large resource consumption and imperfect bird control.Methods of active birds-driving are divided into chemical and physical birds-driving,chemical birds-driving often use bird repellents which make birds disgusted to protect crops.However,most of them are ineffective and costly,and cannot be used on a large scale.Physical birds-driving is simple and effective,so it can be widely applied in solving the problem of agricultural bird damage.The current physical birdsdriving devices usually use static methods to repel birds,but they are single,and birds can easily adapt to them.This method has great effect in the initial period,but as time goes by,the effect will gradually decrease to nothing.Therefore,it is urgent that an accurate and efficient birdsdriving technology is developed to solve the serious problem of agricultural birds damage.Based on the above problems,this paper proposes a real-time bird repellent system based on deep learning.The system uses the unmanned aerial vehicle(UAV)platform to dynamically repel birds based on the detecting results,which efficiently and economically protect agricultural production activities.This paper first selects three most advanced target detection algorithms: Faster R-CNN,SSD and YOLOv5 network.Then all network models are trained and detected through the self-made bird data set,and compared performance from detection performance and detection speed.Finally,the YOLOv5 network with an m AP of 86.4 and a detection speed of 56 frames per second is selected as the target detection algorithm for the bird repellent system,which best meets the experimental requirements.Then design the real-time bird repellent system,which consists of detection device,execution device and communication device.The detection device in the system consists of a camera and a ground station.Its function is to detect bird targets in the field and send bird repelling instructions to the execution device.The execution device is an unmanned aerial vehicle platform that dynamically repels birds according to a predetermined trajectory.The communication device is used to connect and communicate between the UAV platform and the ground station.The communication device is the Raspberry Pi,its function is to enable the UAV platform to receive the bird repellent signal from the ground station and to determine the flight attitude of the UAV platform,and the information is returned to the ground station.Finally,the bird repellent system is tested for the effect of bird repellent.The experimental process of the bird repellent system is to first monitor the protected area through the camera in real time,and determination of bird targets by ground stations.When the target appears,the system will send a bird repellent signal to the UAV platform,so that it can dynamically repel birds according to the prescribed route.After that,the UAV platform will return to the take-off point,at which point the system has completed a working cycle.Then,the experiment was carried out according to the experimental process,and the artificial bird repellent method was used as a control group.The results show that after using the bird repellent system to repel birds,the average time for birds to invade again is more than 30 minutes,and the time gradually increases with the increase in the number of bird repelling system operations,the re-invasion time of birds can reach 48 minutes after the birds are driven away for the third time.But as the control group,the effect of artificial expelled birds is not good,the longest average time for birds to re-invade is 6 minutes,and the shortest is only 3 minutes,and it does not change significantly with the increase in the number of operations.Therefore,it can be obtained from the experimental data that the intelligent bird repellent system based on video detection proposed in this paper can achieve economical and efficient bird repellent effects.