And finally for now, the object detectors are wrapped by a python websocket network wrapper to make it easy for the system to use different types of object detectors. Usually, it's about 1/2 a day for me to write a new python wrapper for a new object detector type. You just need to wrap in the network connection and make it conform to the yolo way of expressing the hits, i.e. the json format that yolo outputs with bounding boxes, class names and confidence level.

What's more, you can even use multiple object detector models in different parts of a single captured image and you can cascade the logic to require multiple object detectors to match for example, or a choice from different object detectors.

It's the perfect anti-poaching system (If I say so myself :) )

Hi all! I'm part of a Pollinator Monitoring Program at California State University, San Marcos which was started by a colleague lecturer of mine who was interested in learning more about the efficacy of pollinator gardens. It grew to include comparing local natural habitat of the Coastal Sage Scrub and I was initially brought on board to assist with data analysis, data management, etc. We then pivoted to the idea of using camera traps and AI for insect detection in place of the in-person monitoring approach (for increasing data and adding a cool tech angle to the effort, given it is of interest to local community partners that have pollinator gardens). 

The group heavily involves students as researchers, and they are instrumental to the projects. We have settled on a combination of video footage and development of deep neural networks using the cloud-hosted video track detection tool, VIAME (developed by Kitware for NOAA Fisheries originally for fish track detection). Students built our first two PICTs (low-cost camera traps), and annotated the data from our pilot study that we are currently starting the process of network development for. Here's a cool pic of the easy-to-use interface that students use when annotating data: 

                                                    Figure 1: VIAME software demonstrating annotation of the track of an insect in the video (red box). Annotations are                                                           done manually to develop a neural network for the automated processing.

The goal of the group's camera trap team is develop a neural network that can track insect pollinators associated with a wide variety of plants, and to use this information to collect large datasets to better understand the pollinator occurrence and activities with local habitats. This ultimately relates to native habitat health and can be used for long-term tracking of changes in the ecosystem, with the idea that knowledge of pollinators may inform resources and conservation managers, as well as local organizations in their land use practices. We ultimately are interested in working with the Kitware folks further to not only develop a robust network (and share broadly of course!), but also to customize the data extraction from automated tracks to include automated species/species group identification and information on interaction rate by those pollinators. We would love any suggestions for appropriate proposals to apply to, as well as any information/suggestions regarding the PICT camera or suggestions on methods. We are looking to include night time data collection at some point as well and are aware the near infrared is advised, but would appreciate any thoughts/advice on that avenue as well. 

We will of course post when we have more results and look forward to hearing more about all the interesting projects happening in this space!

Cheers, 
Liz Ferguson 

I'm looking into writing a sketch for the esp32-cam that can detect pixel changes and take a photo, wish me luck.

One question, does it even need motion detection? What about taking a photo every 5 seconds and sorting the photos afterwards?

Yes. I think this is really the way to go!

Here is another metadata initiative to be aware of. OGC has been developing a standard for describing training datasets for AI/ML image recognition and labeling. The review phase is over and it will become a new standard in the next few weeks. We should consider its adoption when we develop our own training image collections.

Read "OGC Seeks Public Comment on New Standard for Training Data for AI/ML Applications - OGC" on Open Geospatial Consortium

OGC Seeks Public Comment on New Standard for Training Data for AI/ML Applications - OGC

Open Geospatial Consortium

For anyone interested: the GBIF guide Best Practices for Managing and Publishing Camera Trap Data is still open for review and feedback until next week. More info can be found in their news post.

Best,

Max

Hi Sol,

For my research on fish, I had to put together a low-cost camera that could record video for several weeks. Here is the design I came up with

At the time of the paper, I was able to record video for ~12 hours a day at 10 fps and for up to 14 days. With new SD cards now, it is pushed to 21 days. It costs about 600 USD if you build it yourself. If you don't want to make it yourself, there is a company selling it now, but it is much more expensive. The FOV is 110 degrees, so not the 360 that you need, but I think there are ways to make it work (e.g. with the servo motor). 

Happy to chat if you decide to go this route and/or want to brainstorm ideas.

Cheers,

Xavier 

Hi @abra_ash , @MaximilianPink, @Sarita , @Lars_Holst_Hansen. 

I'm looking to train a very compact (TinyML) model for flying pollinator detection on a static background. I hope a network small enough for microcontroller hardware will prove useful for measuring plant-pollinator interactions in the field. 

Presently, I'm gathering a dataset for training using a basic motion-triggered video-capture program on a raspberry pi. This forms a very crude insect camera trap. 

I'm wondering if anyone has any insights on how I might attract pollinators into my camera field of view?  I've done some very elementary reading on bee optical vision and currently trying the following: 

Purple and yellow artifical flowers are placed on a green background, the center of the flowers are lightly painted with a UV (365nm) coat. 

A sugar paste is added to each flower. 

The system is deployed in an inner-city garden (outside my flat), and I regularly see bees attending the flowers nearby. 

Here's a picture of the field of view: 

Does anyone have ideas for how I might maximise insect attraction? I'm particularly interested in what @abra_ash and @tom_august might have to say - are optical methods enough or do we need to add pheremone lures?

Thanks in advance!

Best, 

Ross

Greetings, everyone! I'm thrilled to join this wonderful community. I work as a postdoctoral researcher at MeBioS KU Leuven having recently completed my PhD on "Optical insect identification using Artificial Intelligence". While our lab primarily focuses on agricultural applications, we're also eager to explore biodiversity projects for insect population estimation, which provides crucial insights into our environment's overall health.

Our team has been developing imaging systems that leverage Raspberry Pi's, various camera models, and sticky traps to efficiently identify insects. My expertise lies in computer science and machine learning, and I specialize in building AI classification models for images and wingbeat signals. I've worked as a PhD researcher at a Neurophysiology lab in the past, as well as a Data Scientist at an applied AI company. You can find more about me by checking my website or my linkedin.

Recently, I've created a user-friendly web-app (Streamlit) which is hosted on AWS (FastAPI) that helps entomology experts annotate insect detections to improve our model's predictions. You can find some examples of this work here: [link1] and [link2].  And lastly, for anyone interested in tiling large images for object detection or segmentation purposes in a fast and efficient way, please check my open-source library "plakakia".

I'm truly excited to learn from and collaborate with fellow members of this forum, and I wish you all the best with your work!

Yannis Kalfas

Hi Andrew,

If I understand you correctly, you want to turn on the LEDs when USB power is applied.  The easiest way I can see to do this is to reroute the red wire to USBC VBUS, via an appropriate current limiting resistor.  This bypasses all the electronics in your photo.

You could insert the current limiting resistor in the USB cable for better heat dissipation, or use a DC-DC constant current source instead of a resistor if power consumption is a concern.

Further to @htarold 's excellent suggestion, you can replace that entire PCB with a simple USB breakout board (e.g. USB micro attached below) by removing the red wire and attaching it to VCC on the breakout board, and removing and attaching the black wire to GND. 

Read "SparkFun microB USB Breakout - SparkFun Electronics" on

SparkFun microB USB Breakout - SparkFun Electronics

This simple board breaks out a micro-B USB connector's VCC, GND, ID, D- and D+ pins to a 0.1" pitch header. If you want to add the popular micro-B USB to your project, but don't want to put up with soldering the tiny connector, this is the board for you.