EnsemblePigDet: Ensemble Deep Learning for Accurate Pig Detection

Hanse Ahn,Yongwha Chung,Sungju Lee,Daihee Park,Seungwook Son,Heegon Kim

doi:10.3390/app11125577

Hanse Ahn, Yongwha Chung + Show 4 more

Open Access

https://doi.org/10.3390/app11125577

Copy DOI

Journal: Applied Sciences	Publication Date: Jun 16, 2021
Citations: 18	License type: CC BY 4.0

Affiliation: Korea University, Sangmyung University

Abstract

Automated pig monitoring is important for smart pig farms; thus, several deep-learning-based pig monitoring techniques have been proposed recently. In applying automated pig monitoring techniques to real pig farms, however, practical issues such as detecting pigs from overexposed regions, caused by strong sunlight through a window, should be considered. Another practical issue in applying deep-learning-based techniques to a specific pig monitoring application is the annotation cost for pig data. In this study, we propose a method for managing these two practical issues. Using annotated data obtained from training images without overexposed regions, we first generated augmented data to reduce the effect of overexposure. Then, we trained YOLOv4 with both the annotated and augmented data and combined the test results from two YOLOv4 models in a bounding box level to further improve the detection accuracy. We propose accuracy metrics for pig detection in a closed pig pen to evaluate the accuracy of the detection without box-level annotation. Our experimental results with 216,000 “unseen” test data from overexposed regions in the same pig pen show that the proposed ensemble method can significantly improve the detection accuracy of the baseline YOLOv4, from 79.93% to 94.33%, with additional execution time.

Highlights

Another practical issue in applying deep-learning-based techniques to pig detection is the annotation cost for large-scale training and test data. Traditional metrics, such as average precision (AP)/average recall (AR), which are widely used in COCO [32] and VOC [33], are employed to evaluate the detection accuracy of group-housed pigs
The detection result might not merge correctly, even after applying the model ensemble to the following cases: the same pigs not being detected, A and false positives of model B were created, the detection boxes could not be merged. This was because the false positives of model B were considered as pigs instead of the false negatives of model A, even after applying the model ensemble, due to false positives with high confidence scores To summarize, the detection result might not merge correctly, even after applying the model ensemble to the following cases: the same pigs not being detected, highest confidence score of false positive, and the previous the false highest confidence score of false positive, and the previous case with thecase falsewith negative and negative and false positive occurring at the same time
The proposed method could improve the accuracy of baseline YOLOv4 significantly for pig detection from unseen overexposed regions, the limitations of this study are as follows:

Summary

Introduction

Another practical issue in applying deep-learning-based techniques to pig detection is the annotation cost for large-scale training and test data. Traditional metrics, such as average precision (AP)/average recall (AR), which are widely used in COCO [32] and VOC [33], are employed to evaluate the detection accuracy of group-housed pigs. These require box-level annotation for each pig in an image. For large-scale evaluation, accuracy metrics without box-level annotation are needed

Objectives

Methods

Results

Discussion

Conclusion