Hello all,
I am currently working on a project with the Bosch BME688 and Adafruit ESP32 Huzzah Featherboard (BME688 Development Kit). I began connecting the board as usual to the BME Ai Studio Mobile App. I put it into record data mode, and took the recording of the raw data. The data was saved onto the the SD card that was put inside the board, in the form of .bmerawdata.
I then attempted to graph that data using Matplotlib in Python, but I noticed that the data was very inconsistent and noisy. I am aiming to train my own machine learning learning algorithms, with custom parameters that BME Ai Studio does offer. I have attached a picture of my graphed data below, and as you can see, there is a wide range of gas resistances for each samples. I don't think that the data can be used properly to train the algorithms like I am looking to do.
But, when I put the same data into BME Ai Studio, then the data appears to be filtered and calibrated, as there is no longer drastic sensor drift. I was wondering if any preprocessing is being to the data when uploaded into BME Ai Studio? Is that something which is covered in the documentation and is there somewhere that I can find those functions to implement myself? Or, would it possible to just export the processed data after putting it into BME Ai Studio, but before training the actual algorithm?
Any help or information about this would be much appreciated. As I mentioned before, I have attached an image of the raw data that I graphed myself, with no preprocessing done. If it is possible to just train the algorithm with that data itself, please let me know, but I am not sure about that. Thank you for the advice.
By the way, if you are wondering about the order of the recording, it was done like so:
Normal air, sample 1, normal air, sample 1 variant, normal air, sample 2, normal air, sample 2 variant, normal air, sample 3, normal air, sample 3 variant, normal air, sample 4, normal air, sample 4 variant, normal air. But you can see that the gas resistance fro normal air drifts significantly, and this was something that was not seen in BME Ai Studio, which is what leads to me to wonder if there are any preprocessing steps being done, or library functions being used?
