Towards fruit fly automatic counting: Electronic trap design and long-term feature data acquisition

Bactrocera Dorsalis is a major agricultural insect that severely damages many fruit crops in Southeast Asia. Many integrated pest management (IPM) plans have been implemented to minimize its impact. However, these programs often rely on traditional traps to manually capture and count fruit flies, which is time-consuming and laborious. Current electronic traps have many limitations and face various challenges in their practical implementation. The purpose of this study is to take the first step in developing an electronic trap device capable of automatically collecting the characteristic data of fruit flies that enter the trap over an extended period. Two types of sensors, infrared and sound, were integrated into the trap to gather data on the characteristic behaviors of fruit flies, including trap entry behavior and wingbeat sounds. Analysis of the collected data shows that using infrared data to detect the intrusion of fruit flies into traps achieves an accuracy of 90.72%. Additionally, infrared data revealed several unusual behaviors of the flies while entering the trap, such as entering consecutively, moving around the sensor, and remaining stationary. These behaviors significantly affect the accuracy of counting the number of flies entering the trap but have not been thoroughly analyzed in previous studies. The size of the fruit flies also affected the reliability of the collected data. Moreover, the wingbeat sound data of fruit flies contain distinct frequency characteristics in two ranges: 0–2000 Hz and 5000–8000 Hz, differentiating them from other sound sources. When combining both infrared and sound data, the system could simultaneously detect and count flies entering the trap with an accuracy of 88%. These results suggest that integrating infrared and sound sensors can serve as a new approach for designing traps to monitor fruit fly populations over extended periods because dead flies accumulating at the bottom of the trap do not interfere with the sensors. The use of sound data also opens the possibility of embedding artificial intelligence directly into the trap, enabling it to operate independently, consume minimal energy, and reduce bandwidth usage, etc paving the way for a large-scale fruit fly monitoring system.