2021 Fruit Recognition Progress
Year 2021 has been a wild and crazy year. We worked improving image recognition. We are happy with our results, and hope to further improve in 2022. We compared IT image recognition to manual bud recognition and we got an average range of 94 percent to 96 percent IT bud recognition on the first 15 trees in row 1 of Block 6 as compared to manually counting the buds. This is an important milestone towards validating our bud recognition concept.
According to a consultant with John Deere, the HarvestMoore AI Fruit Finder must be able to successfully recognize and pick at least 90 percent of the tree fruit in order for the device to be viable as a tool that competes with manual labor in the field.
The two photos shown here are of the same tree, a couple of years apart. The first image shows our 2020 attempts at bud and tree fruit recognition with the yellow squares representing AI bud recognition. The second photo is the same tree, using the improved AI bud recognition we tested last spring at pruning season.
As you can see, our approach is beginning to yield results. We couldn’t be more excited to deliver this technology to an industry which has been pummeled by the impacts of the pandemic over the last year.
While we are still a few years away from having a product to offer in the field, this fall’s findings suggest we are on the right track to deliver technology that will address the labor needs of tree fruit and nut orchardists in the U.S.
The first prototype was a modified backhoe that provided HarvestMoore a significant amount of development data. The biggest problem was the arm was too slow and much to heavy. HarvestMoore researched available robotic arms but no arm would meet the needs for operation in the apple orchards. We applied for a National Science Foundation NSF) Startup Grant to develop and build the necessary robotic arm.
In the mean time, we will be getting the necessary tree images in March of 2022.