A Generic Approach to Extend Interpretability of Deep Networks

Silva, Catarina; Morais, António; Ribeiro, Bernardete

doi:10.1007/978-3-031-16474-3_40

Cited by 3 publications

(1 citation statement)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After this study, one can say that D-RISE is the only perturbation-based method founded to explain both model's classification and localization [19]. The other perturbation-based methods designed for image classifiers, such as LIME and RISE, require research efforts to extend them to be used for object detectors [20]. That is why we take D-RISE as a state-of-the-art in order to improve explainability accuracy.…”

Section: Related Workmentioning

confidence: 99%

Explainable AI for Deep Learning Based Potato Leaf Disease Detection

Bengamra,

Zagrouba,

Bigand

2023

2023 IEEE International Conference on Fuzzy Systems (FUZZ)

View full text Add to dashboard Cite

The field of agriculture research has been transformed by deep learning, which has demonstrated impressive capabilities in detecting and classifying plant disease from leaf images. Although several deep learning-based models are proposed, they are considered as blackbox, lack transparency and are notoriously difficult to interpret. Recently, eXplainable Artificial Intelligence (XAI) methods has demonstrated the potential to interpret the model decision-making process through saliency explanations that highlight the most relevant parts of the input image deemed important for predictions. In this work, we proposed a new XAI saliency method for explanation of potato disease detector based on particular perturbations driven by intermediate object detection results. In order to compare our proposed method with the state of the art, qualitative and quantitative experiments are performed for potato leaf disease detector models on PlantDoc dataset.

show abstract

Section: Related Workmentioning

confidence: 99%