Contextual bias in agricultural image networks revealed by neuron concepts

Understanding Contextual Bias in Agricultural CNNs Through Hidden-Neuron Concept Induction

Computer Vision and Pattern RecognitionMachine Learning

Summary

Sometimes, AI image classifiers look at backgrounds instead of the main object to decide what is in a picture. This paper studies what specific features inside an AI’s 'brain' (hidden neurons) respond to when they see these background clues. The authors used a method called Concept Induction to find understandable visual patterns in the AI’s hidden layers from different agricultural image datasets. They found that some neuron responses can be explained by patterns like edges and textures and that this explanation varies by network depth and dataset. This work shows that knowing how these AI neurons work helps complement existing ways to detect AI bias toward backgrounds.

What this means in practice

  • For agricultural ai developers: Identify which internal neural features link to background patterns to improve CNN model robustness on crop image classification.
  • For computer vision engineers: Use Concept Induction to explain hidden-layer activations and diagnose contextual bias in vision models beyond accuracy metrics.

Authors

Abhilekha Dalal, Avishek Das, Samatha Ereshi Akkamahadevi

Abstract

Convolutional Neural Networks (CNNs) can classify images using background-dominated patches that contain little or none of the target object, showing that they exploit contextual information. Previous studies detected this behavior through classification accuracy but say little about what CNN hidden layers represent. We investigate whether Concept Induction can reveal interpretable visual properties associated with hidden-neuron activations on contextual image patches. Using five agricultural datasets with high, moderate, and low contextual bias, we analyze four CNN architectures at three network depths and five patch locations. Quantitative image descriptors of each patch are encoded in a knowledge graph and used as background knowledge for Concept Induction, which derives logical expressions characterizing neuron activation. These expressions are tested on held-out images using the Mann-Whitney U test and Cliff's delta. Of 48,013 tested neuron-feature pairs, 13,164 (27.4%) were validated. Validation was most frequent at intermediate depths (mean 40.2% at mid layers versus 12.0% at deep layers), and no patch location dominated across datasets. Among the tested descriptors, edge, Gabor, and Tamura texture features had the highest validation rates. Stronger contextual bias did not correspond to higher neuron-feature validation rates: moderately biased Sugarcane had the highest rate (40.5%) and highly biased Cassava the lowest (19.2%). These results suggest that behavioral context dependence and the semantic interpretability of hidden neurons capture different aspects of CNN behavior, and that Concept Induction can complement patch-based bias detection.