Results:
The periodic restart has successfully resolved the availability issue, and increasing the confidence threshold to 90% has largely suppressed false detections. However, a significant number of flies and wasps are still being classified as European and Asian hornets. Although it is still early in the season and hornets are currently more focused on gathering protein for their larvae than on seeking carbohydrates, replacing the attractant liquid should be considered to increase the number of insects visiting the stations. This will continue to be monitored throughout July.
Unfortunately, the system’s performance remains below the level required for continuous tracking, particularly when large numbers of insects are present at a station. The overall performance of the stations is still poor, and periods of intense movement frequently reduce the frame rate to just 1-3 FPS. In addition, many images appear blurred, most likely due to the rapid movement of the insects.
While the original design took into account that the stations would be exposed to outdoor conditions, the household rubber bands used to press the attractant bottles against the wooden mounting plate from below were not designed to withstand frequent temperature fluctuations. At two of the four stations, these bands had already failed after only two weeks. Since evaporation of the attractant liquid has so far been minimal and the initial filling is expected to last until the end of the season, a cable tie has now been attached around the bottle neck to prevent the bottle from slipping through the mounting plate.
Hardware Adjustments:
The household rubber bands were replaced with more durable cable ties.
Software Adjustments:
To improve performance, automatic exposure and autofocus will be disabled and replaced with fixed configuration values. In addition, static configuration parameters, such as the GPS location, will be moved to a dedicated config_local.py file, preventing them from being overwritten and requiring reconfiguration whenever minor code changes are deployed. To avoid tracking unnecessarily small insects, such as ants, when camera resolutions change, percentage-based minimum and maximum thresholds for tracking bounding boxes will be implemented. Furthermore, the confidence threshold will be increased to 93% in order to improve detection precision.