Towards reliable recognition for plant diseases and weeds by learning soft probability population
Document Type
Article
Publication Date
3-1-2026
Abstract
Plant diseases and weeds significantly threaten crop yield and food quality, underscoring the necessity for accurate recognition to enable effective management. While deep learning has driven notable progress in this field, a key challenge persists in practical deployment: trained models often yield unreliable predictions when confronted with test samples from new or unknown classes. Current approaches tend to misclassify such samples into one of the known categories, resulting in both prediction errors and missed opportunities to detect novel instances. To overcome this limitation, we introduce the concept of reliable recognition, which assesses whether a trained model’s predictions can be trusted or if human expert intervention is required. This framework encompasses two core tasks: (1) correctly classifying samples into known categories, and (2) effectively distinguishing unknown samples from known ones. To address the second task, we propose a novel scoring function based on a learner model. This learner is trained to mimic the soft probability distribution of the trained classifier’s known-class outputs, thereby enhancing its ability to separate unknown samples. Extensive experiments on four datasets demonstrate the superiority of our approach over state-of-the-art methods. For instance, using ResNet18, our model attains an average open set classification rate (OSCR) of 85.5, outperforming the second-best method, which achieves 84.4. We believe this work establishes a strong foundation for implementing reliable recognition in real-world agricultural systems. To support further research, our code are publicly available at https://github.com/xml94/SPP4OOD .
Keywords
Deep learning, Plant disease recognition, Reliable recognition, Soft probability population, Weed recognition
Publication Title
Engineering Applications of Artificial Intelligence
ISSN
0952-1976
DOI
10.1016/j.engappai.2025.113618
Recommended Citation
Zhou, Maowen; Meng, Yao; Xu, Mingle; Abdullah, Erma Rahayu Mohd Faizal; Sabri, Aznul Qalid Md; and Rahardja, Susanto, "Towards reliable recognition for plant diseases and weeds by learning soft probability population" (2026). Research Publications (2026 to 2030). 1149.
https://knova.um.edu.my/research_publications_2026_2030/1149
Volume
167
Publisher
Elsevier Ltd