Imaging technologies for automated pest detection
Our researchers are developing non-destructive, automated methods for detecting insect pests in fresh produce. These technologies will be particularly helpful to manage biosecurity risks from pests that are difficult to detect through current visual inspection methods. Our efforts are currently focused on detection of key pests of quarantine concern, particularly Queensland fruit fly and Mediterranean fruit fly – species that significantly impact export opportunities for Australia’s $13 billion horticulture industry.
These technologies could be applied in pack houses to identify and remove infested fruit or to enhance inspection methods at the border. Practical imaging technologies that reliably detect infested fruit so that it can be excluded from trade may reduce the need for other more intensive treatment options such as methyl bromide fumigation. When used at the border they could provide an automated more streamlined method that enhances confidence that consignments are pest-free.
Our pest detection technologies can also be used as a research tool to better estimate fruit infestation rates and underpin deeper biosecurity research and innovation.
Optical scanning of fresh produce for grading in a commercial pack house
Optical scanning of fresh produce for grading in a commercial pack house
Optical pest detection system
Many commercial pack houses for horticultural produce use optical scanning as a key component of their quality grading process. These graders are equipped with cameras that utilise near infrared (NIR) spectrum imaging to scan produce for colour, external damage and, in some systems, internal bruising. Optical scanning systems collect hundreds of images per second as the fruit moves along the pack line, and the data is used to grade the produce, and identify fruit that does not meet quality specifications.
Our team is working to develop an NIR imaging system specifically for pest detection that can be integrated into commercial grading technologies. The system uses machine learning to train detection models to identify insect damage in fresh produce. The resulting algorithm can then be used to identify fruit that has a higher risk of pest infestation, enabling the automated grading system to remove this fruit from the packing line.
Optical pest detection prototype
Optical pest detection prototype
Our prototype optical detection system has been trialled through laboratory experiments in a variety of fruits, including cherries, blueberries, apricots, peaches, apples and mangoes. Our initial work has focussed on detecting Queensland fruit fly and Mediterranean fruit fly. The system aims to detect the sting mark made when insects lay eggs in the fruit – a process known as oviposition. To develop an algorithm to detect oviposition stings, we infested fruit in the laboratory with the target pest. We then generated an image library of over 40,000 high-resolution images for each fruit type that we use to train and validate our detection models.
Trial results show that fruit fly oviposition damage is detectable and characteristic when viewed with NIR. Our prototype detection model has achieved an accuracy rate of over 95% in detecting fruit fly damage immediately after oviposition.
Image of cherry showing that fruit fly oviposition damage is very difficult detect through visual inspection.
Some signs of infestation risk in fruit, such as damage from fruit fly stings (oviposition), can be very difficult to detect visually, even for trained inspectors.
Video showing the machine learning process for detecting fruit fly oviposition damage in cherry
A machine learning process is applied to enable fruit fly oviposition damage to be detected through NIR imagery.
The prototype has been designed using similar cameras and lights to those in commercial graders. This will enable the pest detection system to be seamlessly integrated with existing optical grading technologies in pack houses. The research team has secured a patent for this invention and is now working with commercial partners to test the detection system in commercial fruit pack houses.
If the commercial trials are successful, the team will work with market access policy-makers in Australia to prepare technical guidelines that could enable the risk reducing capabilities of commercial optical grading systems to be considered within formal pest risk analysis procedures, or for optical grading to be recognised as a phytosanitary measure.
X-ray imaging
X-ray imaging is one of the most commonly implemented commercial technologies for border security. It is used to detect concealed prohibited items such as weapons, explosives, fresh fruit and other illicit substances.
Our research aims to extend the use of X-ray technology to detect pests within consignments of fresh produce. If integrated into existing border security procedures, the system could expedite inspections, reduce costs, and enhance precision. It could improve overall inspection efficiency and enable biosecurity inspectors to rapidly identify high-risk consignments for more detailed inspection. There may also be potential to deploy pest detecting X-ray technology in commercial pack houses.
This work is a collaborative effort with CSIRO X-ray scientists who have a strong track record in developing and applying leading-edge X-ray imaging technologies. We are using high-resolution cone beam X-ray systems which are well suited to detecting pests and their damage in fresh produce. We are experimenting with both 2D and 3D imaging. We have found that 2D imaging can quickly and effectively capture large pests or infestations and may be more practical for pack house applications. 3D imaging offers higher accuracy in detecting tiny pests and can be implemented with bulk scanning at the borders – though achieving the necessary level of resolution does increase the scanning time. The slower speed for 3D imaging techniques is a disadvantage, however, this can be mitigated through the use of advanced algorithms and hardware devices.
X-ray imagery enables codling moth larvae to be seen in the lower left view of the apple.
Our results show that X-ray imaging can be used to non-destructively detect various internal pests in fruits, nuts and stored product with high accuracy. To date we have tested the system to detect Queensland fruit fly, Codling moth, Carob moth, Carpophilus beetles (dried fruit beetles), and mango seed weevil in their respective hosts. The team have systematically explored the ability of the X-ray system to detect codling moth (CM) in apples, testing fruit infested in the laboratory. Fruits were scanned on different days to capture various larval stages, with some dissected post-scanning to validate image accuracy.
With the integration of AI techniques, our team sees potential for X-ray imaging to evolve into a real-time automated pest detection and quarantine system. As our research progresses, further results will be shared here.
New lines of sight for researchers
The development of imaging technologies for non-destructive detection of insect pests also opens new opportunities for phytosanitary research. For example, NIR imaging could be used to accurately locate and count the number of oviposition sites on an individual fruit. This can provide an alternative method for estimating very low fruit infestation rates, or initial infestation rates in fruit being used in disinfestation studies. Similarly, X-ray technologies can provide a tool to non-destructively study the development and competitive feeding behaviour of pests within fruits and nuts.