Can AI improve insect monitoring? Lessons from Moth Trapping

Tatum Dwyer, UMN Entomology Graduate Student, Eric Yu, UMN Extension IPM Educator, Fei Yang, UMN Extension Corn Entomologist

Every field season, researchers and extension professionals at the University of Minnesota monitor moth flights of several economically important pests, including the western bean cutworm, true armyworm, corn earworm, and European corn borer. Annual moth flight data from previous field seasons can be retrieved from https://vegedge.umn.edu/weekly-moth-flights. Tracking moth flights informs growers and consultants which pests to keep an eye out for and to estimate when to apply control measures through degree-day modeling (read more here: https://vegedge.umn.edu/degree-day-models-select-pests-midwest-region)

One of the most widely used tools for monitoring these insect pests is the blacklight trap. Traditional light traps have been used for centuries to capture nocturnal species. These insects are attracted to the ultraviolet (UV) light atop a funnel and follow their path into a container underneath, where insects are trapped and later identified (Figure 1). These traps can be easily DIY’ed, oftentimes needing only a bucket, a UV light, and battery to be functional. Because many economically important agricultural moths are strongly attracted to UV light, blacklight traps are commonly used to track their emergence and dynamics. However, the biggest disadvantage to using these traps is the requirement of daily visits to the fields to check these traps and the trapped insects must be identified promptly to ascertain which species are active present. Moreover, accurate identification can be challenging because many diagnostic characteristics of moths are located on the wings, which can be easily damaged during trapping and handling. Timely and accurate identification is important because management decisions are often based on the species detected and their abundance. Delays or errors in identification could lead to unnecessary treatment or missed opportunities for effective management.

Figure 1. Traditional blacklight trap at Rosemount Research and Outreach Center (Photo credit: Tatum Dwyer)

Benefits and Limitations with Automated Light Traps

Automated light traps, such as the Trapview AURA trap (Figure 2), have been developed to overcome some limitations associated with traditional light trap by using artificial intelligence (AI) to automatically detect and identify captured insect pests. Like traditional blacklight trap, insects will be attracted to the UV light above a funnel, follow it down, and land on a sticky plate, similar to the yellow sticky cards commonly used for monitoring adult corn rootworm beetles (Figures 2 and 3). A built-in camera inside the trap can take high-resolution pictures of the captured insects and AI-based image recognition software can identify them to species, greatly reducing the need for manual sorting and identification. Because images and identification results are transmitted remotely, users can remotely monitor trap captures and receive real-time updates without making frequent visits to the field. Additionally, the trap automatically advances the sticky plate when it becomes filled with insects, providing a clean surface for continued monitoring. As a result, routine maintenance is minimal, requiring only periodic replacement of the sticky roll after it has been used out.

Figure 2. Initial set up of AURA trap at Rosemount (Photo credit: Tatum Dwyer)

Figure 3. Internal view of the AURA trap (Photo credit: Tatum Dwyer)

Like any monitoring tool, automated traps have both advantages and limitations. Their value depends on the monitoring objectives, target pests, and available resources. Compared with traditional blacklight traps, AURA traps require a higher initial investment, and replacement sticky rolls cost approximately $38 each. During periods of high insect activity, rolls may need to be replaced more frequently, further increasing operating costs. Moreover, each trap only includes one free annual license for the detection and identification of only one target species. If you want to monitor multiple species, you need to purchase separate licenses for each target species, which cost approximately $200 per license per year, further increasing the overall expense. For growers interested in monitoring a single key pest, this may be a practical option. Furthermore, as with any AI-based identification system, the performance of automated traps depends on the accuracy of the detection algorithms. Identification accuracy may vary among target species, although continued advances in AI and image-recognition technology are expected to improve efficiency over time.

Preliminary Results from the University of Minnesota Evaluation

This year, the Yang Lab at the Department of Entomology, University of Minnesota is evaluating the performance of Trapview AURA trap by comparing it with a traditional blacklight trap for monitoring moth flights at the Rosemount Research and Outreach Center. To assess the accuracy of the automated identifications, insects captured by the AURA trap are also manually identified. At this point in the season, only western bean cutworm (WBC) and true armyworm (TAW) have been detected in the traps, and found a generally low pest pressure in the Rosemount area. Preliminary observations indicate some differences between the automated trap and traditional light trap (Figure 4). For WBC, the AURA trap detected a peak flight in early June. However, according to the trap images, these detections were likely celery looper, a minor field and garden pest that is similar in size and coloration to WBC but lacks the diagnostic characteristics used to distinguish WBC for manual identification (Figure 5). In contrast, the traditional blacklight trap recorded a WBC peak flight in mid-June that was not detected by the AURA trap. Discrepancies in the timing of peak moth flights could influence scouting recommendations and the timing of pest management decisions.

Figure 4. Comparison of insect capture between the traditional blacklight trap and AURA trap.

Figure 5. Identifiable characters of Western bean cutworm and Celery looper