Deployment: Analysis¶

As the *metadata_classified.csv file, generated during the classification step, still contains multiple rows for each tracked insect (= track_ID), we will use the provided csv_analysis.py script for metadata post-processing and analysis. The files generated by running the script include a *metadata_classified_top1_final.csv, in which each row corresponds to an individual tracked insect and its classification result with the overall highest probability.

Installation¶

It is assumed that you already followed the instructions in the classification step and successfully ran the classify/predict.py script to classify the cropped insect images and write the classification results to *metadata_classified.csv.

Navigate to the YOLOv5-cls folder, in which you downloaded the insect-detect-ml repo.

Install the required packages by running:

py -m pip install -r insect-detect-ml-main/requirements.txt

Run metadata analysis¶

Navigate to the YOLOv5-cls folder and start the analysis script by running:
```
py insect-detect-ml-main/csv_analysis.py -csv yolov5-master/runs/predict-cls/<name>/results -width 350 -height 200
```
Insert the correct name of your prediction run at <name>. If you used a platform with a different size as the small platform (350x200 mm), change -width and -height to your frame width/hight in mm.
Optional arguments
- -csv path to folder containing metadata .csv file(s)
- -width absolute frame width in mm to calculate true bbox size (default of 1 gives relative bbox size)
- -height absolute frame height in mm to calculate true bbox size (default of 1 gives relative bbox size)
- -min_tracks remove tracking IDs with less than the specified number of images (default = 3)
- -max_tracks remove tracking IDs with more than the specified number of images (default = 1800)

By default, all tracked insects (= track_ID) with less than 3 or more than 1800 images will be removed before saving the *metadata_classified_top1_final.csv. This can exclude many false tracking IDs, e.g. insects moving too fast to be correctly tracked ("jumping" IDs) or objects that are lying on the platform and are incorrectly detected as insects. Depending on the speed and accuracy of the deployed detection model, as well as the respective recording duration, adjusting these thresholds by using the arguments -min_tracks and/or -max_tracks can result in a more accurate estimation of insect abundance/activity (= platform visits).

CSV analysis¶

The csv_analysis.py script will:

Sort metadata:
- read the *metadata_classified.csv file into pandas DataFrame
- calculate the relative (or absolute, if frame width/height in mm is given) bounding box sizes and bbox length (= longer side) + bbox width (= shorter side)
- save metadata sorted by recording ID, tracking ID and timestamp successively (ascending) to *metadata_classified_sorted.csv
Group metadata by tracking ID per recording ID and calculate for each tracking ID:
- number of images per top1 class
- total number of images per tracking ID
- mean classification probability for each top1 class
- weighted mean classification probability for each top1 class
- mean bbox length and bbox width for each top1 class
- save metadata sorted by recording ID, tracking ID (ascending) and weighted probability (descending) successively to *metadata_classified_top1_all.csv
Group metadata by tracking ID per recording ID and calculate for each tracking ID:
- total number of images
- date from first timestamp
- first and last timestamp (start/end time)
- duration [s] (end time - start time)
- mean detection confidence
- for the top1 class with the highest weighted probability:
  - number of images
  - name
  - mean classification probability
  - weighted classification probability
  - mean bounding box length and width
- remove tracking IDs with less or more than the specified number of images
- save metadata calculated for each tracking ID (per recording ID) to *metadata_classified_top1_final.csv
Save some info about the analysis run to *metadata_classified_analysis_info.csv
Create and save some basic plots for a quick first data overview

Overview plots¶

Several plots are generated by the csv_analysis.py script that can give a first overview of the post-processed metadata. For more in-depth statistics, the final .csv file should be analyzed with software such as R + RStudio.

The plot top1_mean_det_conf.png can be used to find cases (e.g. small beetles in the following example) for which the deployed detection model has a low confidence score and additional annotated images could increase model accuracy.

The plot top1_per_rec.png gives a overview of the top1 classes per recording. In the following example, lower numbers of insects at recordings early in the day can be noticed. Also an increase of images classified as dirt (none_dirt) can be observed in later recordings.

The plot imgs_per_track.png gives you information about the distribution of the number of images (= tracking duration) per tracking ID. By default, all tracked insects with less than 3 or more than 1800 images will be removed before saving the final .csv file. You could run the csv_analysis.py script with the arguments -min_tracks 1 and -max_tracks 9999 to plot all tracking IDs and include them in the final .csv file.

To find cases where the accuracy of the classification model could be improved by retraining with additional images added to the basic dataset, you can inspect the plots top1_prob.png and top1_prob_mean.png. In the following example, a relatively low classification probability can be noticed for the classes hfly_myathr (but also only few images), hfly_eristal, beetle and bee_apis. The classified and sorted images in the folder top1_classes should be inspected in cases of such low probabilities to find false classification results.