India is mainly an agriculture based country. Security of agricultural farm is of utmost importance for protecting the produce. Not being able to make the grown crops to the market is another side of the problem. Valuable investments and efforts can be ruined in minutes intentionally or unintentionally by persons or by animals. Introducing machine learning to this problem paves a way to smart agriculture. The proposed system employs Raspberry Pi board to detect any malicious activities or motion in the farm land and triggers the thermal camera to take picture of the scene image. The captured image is fed into the trained CNN (Convolutional Neural Network) model of deep learning. And after detecting the class of the animal, a sound file mapped to that class is played at little higher frequency, so as to scare away the animal. The result shows that the system is 80-95% accurate and 100% consistent for detecting any suspicious movement and to act accordingly.

Keywords : Deep Learning, Convolutional Neural Network, Thermal Cameras, Tensor flow, Fastai, Raspberry Pi, PIR sensor etc.