Real-time animal detection and prevention system for crop fields

Every year, crop damaged by wild animals is dramatically increasing in Sri Lanka. It often poses risks to humans and animals. Since more and more wild animals are causing damage to their cultivation; humans could not tolerate it. Therefore, they require an effective mechanism to overcome this situation. With that background, the objective of this study is to detect wild animals before entering into the crop fields and implementing appropriate scare-away mechanisms in real-time. The presence of the animal will be sent to the farmer via a mobile application. In this study, two Convolutional Neural Network (CNN) classification models have been developed using the transfer learning approach with the VGG-16 as a pre-trained model to detect elephants, wild boars, and buffalos. Both two models were combined and runs on Raspberry pi, which acts as the processing unit for the system, captures the images of animals, and predicts it. Whenever the presence of the animal senses by the thermal sensor which is installed on Arduino, it sends a trigger to capture the image. Based on the prediction sudden flashes of light, ultrasound, and bee sound will be produced to scare away the animals. The mobile application was developed using react native which is used to alert the user about the animal, connected through the Firebase database. The findings of this research indicate that the accuracy rate of the classification model is 77 percentage. This system significantly reduces human-animal conflict in crop fields by automatically implementing scare-away mechanisms based on the prediction.