# An end-to-end convolutional selective autoencoder approach to Soybean Cyst Nematode eggs detection

**Authors:** Adedotun Akintayo, Nigel Lee, Vikas Chawla, Mark Mullaney, Christopher Marett, Asheesh Singh, Arti Singh, Greg Tylka, Baskar Ganapathysubramaniam, Soumik Sarkar  
**arXiv:** [1603.07834](https://arxiv.org/abs/1603.07834)  
**Date:** 2016-03-25  
**Summary page:** https://www.summarizepaper.com/en/paper/1603.07834v1/

## Key points

- SCN infection causes an annual loss of approximately $1 billion in the United States
- Current identification methods rely on manual microscopy, which is labor-intensive and time-consuming
- A novel approach for detecting SCN eggs in soil samples using deep convolutional networks is proposed
- The framework uses a selective autoencoder trained on manually labeled microscopic images to suppress undesired parts of an image frame while allowing the desired parts, resulting in efficient object detection
- The problem is challenging due to the presence of non-egg particles (disturbances) that are very similar to SCN eggs in shape, pose, and illumination
- The selective autoencoder is trained to learn unique features related to the invariant shapes and sizes of the SCN eggs without handcrafting
- After training, a composite non-maximum suppression and differencing is applied at the post-processing stage for automated high-throughput egg detection
- Experiments with real-world data demonstrate the efficacy of the framework
- The proposed approach has potential for expediting SCN egg detection and reducing losses due to nematode infections on soybean plants.

## Summary

This paper presents a novel approach for detecting Soybean Cyst Nematode (SCN) eggs in soil samples using deep convolutional networks. SCN infection causes an annual loss of approximately $1 billion in the United States, and current identification methods rely on manual microscopy, which is labor-intensive and time-consuming. The proposed framework uses a selective autoencoder trained on manually labeled microscopic images to suppress undesired parts of an image frame while allowing the desired parts, resulting in efficient object detection. The problem is particularly challenging due to the presence of non-egg particles (disturbances) that are very similar to SCN eggs in shape, pose, and illumination. The selective autoencoder is trained to learn unique features related to the invariant shapes and sizes of the SCN eggs without handcrafting. After training, a composite non-maximum suppression and differencing is applied at the post-processing stage for automated high-throughput egg detection. The efficacy of the framework is demonstrated through experiments with real-world data, showing its potential for expediting SCN egg detection and reducing losses due to nematode infections on soybean plants.