[Paper Review] InsectMamba: Insect Pest Classification with State Space Model

Qianning Wang, Chenglin Wang+2 more|arXiv (Cornell University)|Apr 4, 2024

Species Distribution and Climate Change8 citations

TL;DR

InsectMamba fuses State Space Models with CNNs, MSA, and MLPs in Mix-SSM blocks plus a selective module to improve insect pest classification across five datasets, achieving superior results over strong baselines.

ABSTRACT

The classification of insect pests is a critical task in agricultural technology, vital for ensuring food security and environmental sustainability. However, the complexity of pest identification, due to factors like high camouflage and species diversity, poses significant obstacles. Existing methods struggle with the fine-grained feature extraction needed to distinguish between closely related pest species. Although recent advancements have utilized modified network structures and combined deep learning approaches to improve accuracy, challenges persist due to the similarity between pests and their surroundings. To address this problem, we introduce InsectMamba, a novel approach that integrates State Space Models (SSMs), Convolutional Neural Networks (CNNs), Multi-Head Self-Attention mechanism (MSA), and Multilayer Perceptrons (MLPs) within Mix-SSM blocks. This integration facilitates the extraction of comprehensive visual features by leveraging the strengths of each encoding strategy. A selective module is also proposed to adaptively aggregate these features, enhancing the model's ability to discern pest characteristics. InsectMamba was evaluated against strong competitors across five insect pest classification datasets. The results demonstrate its superior performance and verify the significance of each model component by an ablation study.

Motivation & Objective

• Address camouflage and high species diversity in insect pest classification.
• Leverage complementary visual encoding strategies to extract both local and global features.
• Introduce Mix-SSM blocks that integrate SSM, CNN, MSA, and MLP.
• Develop an adaptive Selective Module to fuse features from different encoding strategies.
• Demonstrate superior performance over strong baselines on five insect pest datasets.

Proposed method

• Propose Mix-SSM blocks that integrate SSM, CNN, MLP, and MSA for feature extraction.
• Introduce a Selective Module that adaptively weights and aggregates features from multiple encodings.
• Represent images as patches, apply PatchEmbed, and process through stacked Mix-SSM Blocks with PatchMerging.
• Compute channel-wise and spatial attention within the Selective Module via global pooling and an MLP, producing per-channel weights for each encoding strategy.
• Train and evaluate on five insect pest datasets, comparing against ResNet, DeiT, Swin, Vmamba, and baselines.
• Conduct ablation studies to verify the contribution of CNN, MLP, and MSA components.

Experimental results

Research questions

• RQ1 Can Mix-SSM blocks effectively integrate SSM, CNN, MSA, and MLP features for insect pest classification?
• RQ2 Does the Selective Module improve adaptive fusion of multi-encoding features across diverse datasets?
• RQ3 What is the impact of individual components (CNN, MLP, MSA) on overall performance and ablation outcomes?

Key findings

• InsectMamba achieves state-of-the-art performance across five insect pest datasets, outperforming strong baselines in ACC, Prec, Rec, and F1.
• On Farm Insects, InsectMamba attains ACC 0.66 and F1 0.65, surpassing the next best model Swin-B by about 4 percentage points in ACC.
• On Agricultural Pests, InsectMamba reaches ACC 0.91 and F1 0.91, beating Vmamba-B by about 2 percentage points in ACC.
• On Insect Recognition, InsectMamba achieves ACC 0.86 and F1 0.86, outperforming strong baselines.
• On Forestry Pest Identification, InsectMamba achieves ACC 0.94 and F1 0.94, the best among listed methods.
• On IP102, InsectMamba attains ACC 0.43 and F1 0.37, improving over the previous best (Swin-B) with ACC 0.39.