In Japan, damage to crops by wild boars and other wild animals is becoming increasingly serious due to the aging of hunters and the lack of effective trapping devices, causing significant damage to farmers. As a countermeasure, farmers are using traps such abox type trap, fence trap and wire trap, however these methods are essentially manual thus limited in efficiency to responds against the crop harmful animals. Therefore, a recognition algorithm is needed for recognizing wild boar presence and capture them automatically by Artificial intelligence. Therefore, in our previous study, a recognition algorithm was used to develop to recognize boars in box traps and a digitally controllable capturing device, and these were introduced into a microcontroller; however, there were limitations in accuracy and processing efficiency. Therefore, the objective of this study is to select the best AI model in terms of accuracy and speed on a microcontroller. A new dataset was created using data collected in the experimental site, and AI models were trained using two different frameworks: Pytorch for flexible processing and Darknet for lightweight and fast inference. The data types were then converted using quantization techniques and implemented on a microcontroller to compare and verify each data type and framework. This comparison allowed us to select the optimal AI model on the microcontroller.