AI detecting ripe fruits with human-level accuracy

See Tree is one of the leading companies in agricultural tech field. The number one priority of See Tree is to oversee their fruiting status of trees in order to increase their productivity. This calls for a device for automated detection of ripe fruits using machine learning techniques. To have better, faster results and to approach this task with all the tools that are needed, See Tree partnered with Ciklum, a company whose task is to help other companies in creating something new with their own specialists.
In this case Ciklum provided See Tree with a team of  skilled R&D experts that had to find a solution to these challenges:

  • Small dataset (~500 annotated photos). It was impossible to collect fresh data because oranges were not in season.
  • High level of noise in dataset annotation. Annotations to some images didn’t contain required data. Some bounding boxes were not precise enough.
  • Dataset contained photo duplicates with different labelling.
  • High occlusion level in some pictures, made it difficult to separate instances.

Data preparation 

To prepare data that is required for this task, Ciklum’s team duplicated photos and images of unlabled oranges or invalid polygons. Those images were filtered out of the dataset. The resulting dataset was divided into sets of train, validation and test,  afterwards, stratified by the wide variety of oranges located on the picture.

 

 

 

 

 

Model architecture 

The satisfactory end result became acquired with the Faster R-CNN architecture family.This pattern became a modern

 

method for resolving object detection task. ” Faster R-CNN is a two-stage object detection system in which the first

stage generates a sparse set of candidate object locations and the second stage classifies each candidate location as one of the foreground classes or as background using a convolutional neural network.”

 

 

 

 

 

 

 

 

Result

Ciklum’s and See Tree’s prototype had these advantages:

  • able to detect ripe fruits with high accuracy, even if occluded;
  • although built to detect oranges, it can be easily re-trained to detect other kinds of fruits;
  • outperforms humans in most cases.

 

 

 

 

 

All references were taken from:

https://www.ciklum.com/case-studies/seetree/

Radomyr Kostetskyi

The following two tabs change content below.
Avatar

Kostetskyi Radomyr

Avatar

Latest posts by Kostetskyi Radomyr (see all)

Leave a Reply