You can view the full text of this article for free using the link below.

Title: Multi-stream aggregation network for fine-grained crop pests and diseases image recognition

Authors: Xuebo Jin; Zhi Tao; Jianlei Kong

Addresses: Beijing Technology and Business University, Beijing, China ' Beijing Technology and Business University, Beijing, China ' Beijing Technology and Business University, Beijing, China

Abstract: Pests and disease recognition can be considered as Fine-Grained Visual Classification (FGVC) problems, suffering low inter-class discrepancy and high intra-class variances from the sub-categories, which is more challenging than common basic-level category classification dependent on traditional Deep Neural Networks (DNNs). To encourage further progress in challenging realistic agricultural conditions, this paper presents a realistic CropDP181 data set with 181 categories and fine-grained multi-stream aggregation network with models transferred named as MSA-NET (Multi-Stream Aggregation Network) for fine-grained species recognition based on fusion idea. The novel MSA-NET model combines ResNet, NTS-Net (Navigator-Teacher-Scrutiniser Network), and FAST-MPN-COV (Towards Faster Training of Global Covariance Pooling Network) trained by a multi-stream feature extractor to exploit the high-dimensional feature maps representing discriminative and non-discriminative parts as well as interclass variances. Finally, a fusion module equipped with NetVLAD (Network Vector of Locally Aggregated Descriptors) layer is developed to fuse different components model as a unified probability representation for the ultimate fine-grained recognition. The MSA-NET model achieves competitive results in fine-grained pests and disease recognition outperforming state-of-the-art methods.

Keywords: crop pests and diseases; fine-grained visual classification; multi-stream neural networks; NetVLAD aggregation.

DOI: 10.1504/IJCCPS.2021.113105

International Journal of Cybernetics and Cyber-Physical Systems, 2021 Vol.1 No.1, pp.52 - 67

Received: 15 Jan 2020
Accepted: 28 May 2020

Published online: 31 Jan 2021 *

Full-text access for editors Access for subscribers Free access Comment on this article