The goal of this website is to allow automated identification of species of seed beetle that are regularly intercepted at U.S. Ports of Entry. We are using Artificial Intelligence models and Machine Learning algorithms trained on thousands of identified specimens. We cannot guarantee accuracy for every specimen, but have achieved high performance for most species. Associated Fact Sheets and identification guides can be consulted to confirm identification, and USDA employees can have specimen identifications confirmed from uploaded photographs.
This material was made possible, in part, by a Cooperative Agreement from the United States Department of Agriculture’s Animal and Plant Health Inspection Service (APHIS). It may not necessarily express APHIS’ views.
Our AI models are trained to recognize the species listed to the right. More common species with more data tend to be easier for our program to recognize while species with less samples are harder to identify. As more specimens are identified using our app, our dataset grows and our models improve their accuracy on the most common species uploaded.
•Acanthoscelides griseolus
•Acanthoscelides obtectus
•Acanthoscelides pallidipennis
•Acanthoscelides zeteki
•Algarobius prosopis
•Bruchidius uberatus
•Bruchus affinis
•Bruchus ermaginatus
•Bruchus lentis
•Bruchus rufimanus
•Bruchus signaticornis
•Callosobruchus chinensis
•Callosobruchus phaseoli
•Callosobruchus rhodesianus
•Callosobruchus theobromae
•Caryobruchus gleditsiae
•Megabruchidius tonkineus
•Mimosestes mimosae
•Pachymerus nucleorum
•Specularius impressithorax
•Stator pruininus
•Zabrotes subfasciatus
•Acanthoscelides argillaceus
•Acanthoscelides macrophthalmus
•Acanthoscelides obvelatus
•Acanthoscelides quadridentatus
•Algarobius bottimeri
•Bruchidius badjii
•Bruchidius villosus
•Bruchus brachialis
•Bruchus ervi
•Bruchus pisorum
•Bruchus rufipes
•Callosobruchus analis
•Callosobruchus maculatus
•Callosobruchus pulcher
•Callosobruchus subinnotatus
•Caryedon gonagra
•Decellebruchus atrolineatus
•Mimosestes amicus
•Mimosestes nubigens
•Pseudopachymerina spinipes
•Stator limbatus
•Stator vachelliae
Our program focuses on quick, easy, and reliable Bruchinae classification. Our AI models provide their best results when the provided images are consistent with the images used in training. The tool tips on our upload page provide examples and explanations of how to create a consistent image for our program to evaluate. Cropping the image is not necessary as long as the beetle appears in a similar manner to the image to the left.
Aligning the beetle properly under the microscope and ensuring that the specimen is clearly separated from the microscope's vignette helps the model by preventing excess dark spots in the cropped image. Key features should be in focus in the image such as the elytra for the dorsal view and the head for the frontal view. If unsure about what to focus the microscope on, try and get as much of the beetle in focus as possible for the best results. The image on the right shows an improperly aligned specimen which is very close to the vignette and mostly out of focus.
Exposing the specimen properly can also help with improving the classification results. Under and overexposed beetles may cause their coloration to be harder to distinguish. Our image preprocessing will help reduce differences caused by exposure levels and enhance the colors provided, but a well exposed image will always provide better results.
Our preprocessing utilizes a simple object detection model to crop images down so that the beetle fills the whole image. We then use some color and feature enhancing techniques before feeding the images to their respective models. We use Resnet50 models from pyTorch and our database of over 10,000 images to create strong image recognition models trained purely for recognizing Bruchinae species. Our program will save your previous 50 classifications in your history. Our admin can review and verify your classifications which will then be added to our training data to improve our model.
This website was developed by a small group under the supervision of Dr. Geoffrey Morse at the University of San Diego. Each role is listed
below with the respective contributors.
Bruchinae expert and database creator: Dr. Geoffrey Morse
Main web and AI developers: Joe Cox and Thomas McKeown
AI developer: Daniel Daugbjerg
Website developers: Joe Boxberger, Sara Evans, and Audrey Krishnadasan
