Knowing about the presence of certain type of insects at an early stage is crucial to prevent a plague that can ruin you crops, especially inside greenhouses.
Usually the task is taken care by a human in the field to detect the presence of those pests to implement the adequate prevention mesures.
This is a time consuming and subjective approach that I believe could be solved by Farmbot, if you incorporate sensors capable of identifying insect type and quantity, logging and notifying them for human action(for now).
I believe you could use the technology described here. It’s made with cheap sensors and can even distinguish if the insect is male or female.
That technology sounds fascinating, and could be a great DIY mod or add-on. We have priorities that preclude us from working on it, but you are welcome to work on it and implement it yourself or with other members of the community. Share your results if you do!
There are many different insects identify techniques and equipment that can be used to detect insects in the air, on plants, and even on and beneath the soil. Having pests removed is a costly thing, but the costs of not finding an infestation early can create a bigger issue. Because pests can create diseases very quickly if left to their own devices, the greater the size of the infestation, the more costs you will incur. So using various techniques or getting a pest exterminator in Fairfield County CT, both will help to remove all pests which can lead to your property damage in future.
First, the lack of effective sensors has made data collection difficult. Most efforts to collect data have used acoustic microphones2-5. Such devices are extremely sensitive to wind noise and ambient noise in the environment, resulting in very sparse and low-quality data.
Second, compounding these data quality issues is the fact that many researchers have attempted to learn very complicated classification models, especially neural networks6-8. Attempting to learn complicated classification models, with mere tens of examples, is a recipe for over-fitting.
Something that could be done is to mount a UV light near the camera, and then take images at night. The insects either glow, or they appear as black spots on slightly purple leaves. check this out. https://www.nightsea.com/galleries/caterpillar-fluorescence/
Computer vision could then be used to identify the insects.
This won’t be foolproof, as some of the insects hide on the underside of the leaves, but it’s something.
Using this system would work quite well probably. Although I would use it to generated email to the human with a picture when the computer’s ability to improve accuracy on detecting bugs great. But in the meantime at least you’re farmbot will be able to say “look look Boss I found something that looks like a bug” so if you added in a system that identifies there is no wind and the camera stops near plants looking for movement it could also warn you of small parasites small enough that most of the cameras won’t pick up very well but the software definitely can see movement all over the place when you have little micro insects moving about all the time
I’ve often wondered if there’s some sort of open source “Facial Recognition” software that could be modified/incorporated for use with the camera to detect undesirable insect pests.
In the end, I figure it would be easier and cheaper to just keep an eye out myself.
What would be nice is to have an infrared sensor to detect bigger pests like deer, rabbits, raccoons, etc. that would either sound an alarm or shoot a water stream to chase them off.
I barely can get recognition of fruit and vegetables to work, vs the challenges of moving creatures and many times very very small insects… that said one could derive evidence of insect infestation , leaves eaten etc…
Is it a data volume problem? When I did a bit of work on image recognition the biggest issue I had was getting enough images. There are some free online image libraries we used but still it was difficult to get matches on the last 10-15%. Given you’re using a 3D scanner I expect there are even less 3D image libraries out there.
We needed more data i.e. images to train the neural net. It got better the more images we trained it in. There were also problems with shadowing in images, which we solved by running the images through some “de-shadowing” algorithms.
Yeah I agree 3D mage recognition will be better in the future than 2D but it seems a long way off.
@whitecaps,
I mentioned this in another thread, you may be interested in. Dual camera with “AI processor” (MyriadX from Intel) and raspberry Pi for 3D processing:
