24/7 Raspberry Pi/birdNET monitoring (4 Viewers)

[LittleBitOfBreadNoCheese]

What do you mean drop off? Leaving it at a remote location?

yes

 

[Jon.Bryant]

but would love to record 24/7 in a waterproof device

If there is no WiFi/internet connection on power-up, then the PUC automatically goes into SD Recording Mode. It will store the audio recordings (WAV or FLAC) on the SD card, as well as writing sensor data to the PUC_Data.csv file.

Probably stating the obvious, but I think the PUC is really designed around real time analysis and hence as a base station with a WiFi or Internet connection.

If all it does is record audio to an SD card when there is no connection - which would then presumably need to be fed back through an analysis programme (or checked manually), then there is no difference to passive recorders already out there - I think the Wildlife Acoustics recorders have ruled the passive recorder market for a while and their SM4 can be left in the field independently for a very long time (it is powered by 4 DD batteries and has dual SD card slots). The Micro field life is less, but it is reasonably priced and the mic is pretty sensitive to frequencies common for many passerines.

Not sure what the PUC does, but a limit passive recorders is that they are generally 16bit 44KHz to preserve battery life. You also need to be careful with placement and gain levels, as close vocals will often clip.

 

[Jon.Bryant]

Now if the PUC could do analysis offline that would be amazing. Imagine being able to interrogate a station and get a list of all the species it had detected, plus the linked sound files for confirmation.

 

[GreenRooky]

Now if the PUC could do analysis offline that would be amazing. Imagine being able to interrogate a station and get a list of all the species it had detected, plus the linked sound files for confirmation.

Not sure if I am missing something but this is what you can do isn't it? Once back within range of WiFi it uploads everything to the dashboard.

I was thinking about this but it if you were very keen, there is no reason you couldn't have a powerbank running the PUC and a 4G dongle and have it running in the field in real time but I think it would chew through data

 

[LittleBitOfBreadNoCheese]

Not sure if I am missing something but this is what you can do isn't it? Once back within range of WiFi it uploads everything to the dashboard.

I was thinking about this but it if you were very keen, there is no reason you couldn't have a powerbank running the PUC and a 4G dongle and have it running in the field in real time but I think it would chew through da

That is a good description of what the app does if you take it outdoors but have no data. It does detections in realtime (minus species icons for birds you have only heard outside) and when you get back in they can all be uploaded to the site map quite quickly with accurate map points too. However I assume the PUC version when you get back on the wifi has to stop to do all the detection work first. If you are only going to interrogate on retrieval there isn't much point in detecting in realtime.

wrt your 2nd paragraph the big question is how much power it would consume, no doubt Tim evaluated this at design time. There is a huge difference in requirements to run the neural net 24/7.

 

[GreenRooky]

That is a good description of what the app does if you take it outdoors but have no data. It does detections in realtime (minus species icons for birds you have only heard outside) and when you get back in they can all be uploaded to the site map quite quickly with accurate map points too. However I assume the PUC version when you get back on the wifi has to stop to do all the detection work first. If you are only going to interrogate on retrieval there isn't much point in detecting in realtime.

wrt your 2nd paragraph the big question is how much power it would consume, no doubt Tim evaluated this at design time. There is a huge difference in requirements to run the neural net 24/7.

Which app?

 

[GreenRooky]

Does anyone know what the different colour outlines to the bird icons mean in the birdweather app?

 

[LittleBitOfBreadNoCheese]

Which app?

The BirdWeather app. It has one mode to control a PUC but (iOS only I think) can run a standalone BW station.

 

[GreenRooky]

Ah OK, I'm on android and couldn't see that functionality within the app

 

[Jon.Bryant]

That is a good description of what the app does if you take it outdoors but have no data. It does detections in realtime (minus species icons for birds you have only heard outside) and when you get back in they can all be uploaded to the site map quite quickly with accurate map points too. However I assume the PUC version when you get back on the wifi has to stop to do all the detection work first. If you are only going to interrogate on retrieval there isn't much point in detecting in realtime

I am confused by this statement, which seems a little contrary to the following.

If there is no WiFi/internet connection on power-up, then the PUC automatically goes into SD Recording Mode. It will store the audio recordings (WAV or FLAC) on the SD card, as well as writing sensor data to the PUC_Data.csv file.

Unless you mean that the recorder does not record to SD continously, but only at some noise threshold (a detection?). If the latter then detecting bird sound is generally rather flaky - very hard to differentiate ticks and chats of bird calls from ambient noises and mic pops and clicks. Generally works much better with algorithms looking for song (so many noises, with a max gap of 1/x of a second etc.). I tried using the Wildlife Acoustic detection algorithm for nocmigging and if you set it to detect short sounds, the results are lousy.

Also when you say accurately record location, what do you mean? Presumably the location is taken from a phone tether, or does the PUC have its own GPS. Sound files do not have position properties, so if you were ‘moving’ with the PUC (and it has GPS), you would need to record a KMZ file (or similar track file) and then use the time stamp of the sound file (plus time offset in a long recording) to calculate the position from the track fole - and then create a small database of detections and locations. Some of the Wildlife Acoustics recorders tether to a phone for initial setup including location setting, and this data is written to a dat file - but this is a single position set once in deployment. I think there is some clever way you can do bat transects with the WA recorders and match detections with locations (presumably provided by a phone tether), but this option is not available for birds. I think this is because bat are much easier to isolate:detect (given the high frequency) and that recording full spectrum recordings at ultra high frequencies would eat memory, so running a buffer and only recording detections makes sense.

Does anyone have first hand experience of how the PUC works and detections are made when using the device offline?

 

[LittleBitOfBreadNoCheese]

I am confused by this statement, which seems a little contrary to the following.

Unless you mean that the recorder does not record to SD continously, but only at some noise threshold (a detection?). If the latter then detecting bird sound is generally rather flaky - very hard to differentiate ticks and chats of bird calls from ambient noises and mic pops and clicks. Generally works much better with algorithms looking for song (so many noises, with a max gap of 1/x of a second etc.). I tried using the Wildlife Acoustic detection algorithm for nocmigging and if you set it to detect short sounds, the results are lousy.

Also when you say accurately record location, what do you mean? Presumably the location is taken from a phone tether, or does the PUC have its own GPS. Sound files do not have position properties, so if you were ‘moving’ with the PUC (and it has GPS), you would need to record a KMZ file (or similar track file) and then use the time stamp of the sound file (plus time offset in a long recording) to calculate the position from the track fole - and then create a small database of detections and locations. Some of the Wildlife Acoustics recorders tether to a phone for initial setup including location setting, and this data is written to a dat file - but this is a single position set once in deployment. I think there is some clever way you can do bat transects with the WA recorders and match detections with locations (presumably provided by a phone tether), but this option is not available for birds. I think this is because bat are much easier to isolate:detect (given the high frequency) and that recording full spectrum recordings at ultra high frequencies would eat memory, so running a buffer and only recording detections makes sense.

Does anyone have first hand experience of how the PUC works and detections are made when using the device offline?

I haven't tried it (not that bothered about using it that way) but the PUCs have onboard GPS. The data is stored in FLAC files with a meta data file containing the GPS derived lat and long for each FLAC. This is covered in the app's FAQ under 'PUC SD Card'. My guess at this stage is they are 'pseudo-detections' as one might call them. But the PUC is fiddly enough plugged in so I will let someone else investigate...

 

[GreenRooky]

Been using my new PUC a few days now. I was getting a bunch of Hawfinch reports when it was a Robin, tweaking the probability to 7% and the confidence to 55% has got rid of these and a lot of other detections. I guess the correct way to deal with it but would be up report the erroneous identifications as bad to improve the algorithm but currently the Android app doesn't support playback so would need to be done via the website which is clunky, android playback is on the road map though so hopefully this will be easier soon.

I'm running it on batteries, even with 2800mah NiMh cells they are only lasting around 30 hours although it is getting close to freezing at night currently, so I need to find a better way to power it outdoors.

Despite the vast number of detections being robins and blue tits, it has also captured night time fly over coot, wigeon, teal, curlew and oysyecatcher. It was these types of things I was hoping for. Sadly it didn't pick up the tawny owl that was calling so loudly that I could hear it indoors over the TV, possibly the changes to the settings I made caused this.

It has sat out in very heavy prolonged rain and had no issue which is reassuring.

Once I figure out a better power options, probably a power bank in a waterproof box I will deploy it remotely, a dog walker the other day told me where he believes there are willow tit or marsh tit near by and I have yet to be able to locate them on several visits, hopefully this would allow me to confirm that are at least in the area and worth me putting in more time looking.

 

[LittleBitOfBreadNoCheese]

Been using my new PUC a few days now. I was getting a bunch of Hawfinch reports when it was a Robin, tweaking the probability to 7% and the confidence to 55% has got rid of these and a lot of other detections. I guess the correct way to deal with it but would be up report the erroneous identifications as bad to improve the algorithm but currently the Android app doesn't support playback so would need to be done via the website which is clunky, android playback is on the road map though so hopefully this will be easier soon.

I'm running it on batteries, even with 2800mah NiMh cells they are only lasting around 30 hours although it is getting close to freezing at night currently, so I need to find a better way to power it outdoors.

Despite the vast number of detections being robins and blue tits, it has also captured night time fly over coot, wigeon, teal, curlew and oysyecatcher. It was these types of things I was hoping for. Sadly it didn't pick up the tawny owl that was calling so loudly that I could hear it indoors over the TV, possibly the changes to the settings I made caused this.

It has sat out in very heavy prolonged rain and had no issue which is reassuring.

Once I figure out a better power options, probably a power bank in a waterproof box I will deploy it remotely, a dog walker the other day told me where he believes there are willow tit or marsh tit near by and I have yet to be able to locate them on several visits, hopefully this would allow me to confirm that are at least in the area and worth me putting in more time looking.

But is Hawfinch a plausible species for you because it isn't at all for me hence my annoyance at the reports. By contrast if you use birdNET-Analyzer you can feed it the lat and long (still not convinced BW does this and/or actions it properly) and if that doesn't work you can edit the species list and just remove it. The other problem report is Green Sandpiper but that is theoretically possible in some seasons so I wouldn't want to remove it completely.

 

[GreenRooky]

Hawfinch as a flyover possible but very unlikely, sat singing in my garden daily almost impossible. This is why I have adjusted the probability and confidence levels, the hawfinch was always getting reported with very low confidence. This does mean something rare could get missed but I would rather than than many false positives. I think it also shows the challenge of relying on ebird for the probability, if you live in an area with very low reporting, I have to assume it has a very weak dataset to work off.

My understanding from the FAQ is that location is used for determining probability.

 

[LittleBitOfBreadNoCheese]

Hawfinch as a flyover possible but very unlikely, sat singing in my garden daily almost impossible. This is why I have adjusted the probability and confidence levels, the hawfinch was always getting reported with very low confidence. This does mean something rare could get missed but I would rather than than many false positives. I think it also shows the challenge of relying on ebird for the probability, if you live in an area with very low reporting, I have to assume it has a very weak dataset to work off.

My understanding from the FAQ is that location is used for determining probability.

Well I use BirdTrack so all bets are off if they rely on other databases. But the last Clyde region annual bird report has zero sightings of Hawfinch.

 

[GreenRooky]

I think one of the current problems with BirdWeather is lack of documentation unless I have missed any documentation beyond the FAQ or there is somewhere else it is discussed? Hopefully it will improve with time and as more users get thier PUC's

 

[LittleBitOfBreadNoCheese]

I think one of the current problems with BirdWeather is lack of documentation unless I have missed any documentation beyond the FAQ or there is somewhere else it is discussed? Hopefully it will improve with time and as more users get thier PUC's

Not yet, maybe a keen customer will start a wiki?

 

[Jon.Bryant]

even with 2800mah NiMh cells they are only lasting around 30 hours although it is getting close to freezing at night currently, so I need to find a better way to power it outdoors.

I think a rechargeable power bank will also be impacted by cold weather. A bit expensive (and perhaps not that environmentally sound), but I imagine the single use Energizer ultimate lithium batteries are the way to go in cold weather. They work well down to -20c.

 

[MiddleRiver]

I think a rechargeable power bank will also be impacted by cold weather. A bit expensive (and perhaps not that environmentally sound), but I imagine the single use Energizer ultimate lithium batteries are the way to go in cold weather. They work well down to -20c.

Rechargeable LiOn

 

[GreenRooky]

Yes Lion look like the have a better life
They are also promising a firmware update to improve battery life but I think the might involve changing the recording profile/ frequency

I think a 30,000mah powerbank would cover my needs

Typical battery life on current firmware (running as a live station, posting every soundscape for analysis):

• Amazon Basics Alkaline ... ~30hrs
• Energizer Lithium ... ~65hrs
• CZVV Lithium Rechargeable ... ~48hrs
• NIMH ... ~40hrs