Skip to content

README6.md

Latest commit

 

History

History
109 lines (71 loc) · 3.89 KB

README6.md

File metadata and controls

109 lines (71 loc) · 3.89 KB

Coyote Project

Team Coyote2 : Deep Learning

Contents

  1. Team
  2. Overview
  3. Research problem statements
  4. Research novelty
  5. Requirements
  6. Technology Stack
  7. Deep Learning

Team

Name University Department Email Contact
Yejin Lee Hallym University Dept. of Computer Science leeye0616@naver.com https://github.com/yetniek
Heesun Jung Hallym University Dept. of Computer Science glee623@naver.com https://github.com/glee623
Youngbin Kim Kwangwoon University Dept. of Computer Information binny9904@naver.com https://github.com/0binn
BoKyung Kwon Kwangwoon University Dept. of Computer Information bbo1209@naver.com https://github.com/doomdabo
Jihyun Park Jeju National University Dept. of Computer Science & Statistics mmmszip@gmail.com https://github.com/mmmtobezip
Griffin Pegg Purdue University Dept. of Computer and Information Technology pegge@purdue.edu https://github.com/coyotehowls

Overview

overview of coyote2

Research problem statements

The attacks on livestock, human, and crops by coyotes are occurring over the United States, while traditional simple management such as public education about the method of avoiding coyotes and coyote hunting contests to reduce their numbers are executed. There are not sufficient cases of technical approaches or research about the damage to coyotes.

Research novelty

The method of coyote howling sound classification using Convolutional Neural Network (CNN) to reduce the damage of coyotes is needed. This paper suggests using a network connection in order to prevent the damage by informing the neighborhood farms when coyotes appear and chasing coyotes through a coyote alert system. It is expected that additional technical approach to current coyote damage prevention can improve the accuracy and make the previous management more practical.

Requirements

version

Python 3.7 ~ 3.9

Colab

Librosa

Configuration

conda install -c conda-forge pyngrok

file structure

C:.
│ base_line.ipynb
│ make_mel_s.ipynb
│ move_wav_file.ipynb
│ train.csv
│ valid.csv
│
├─dataset
│ ├─train
│ └─valid
├─dataset_all
└─mel_spectrogram ├─train_mel
└─valid_mel

Technology Stack

Deep Learning

Dataset

There are a total of 1,160 training dataset. It consisted of 586 coyotes, 480 dogs, and 94 chickens. And the test data set is a total of 280 sheets. It consisted of 134 coyotes, 117 dogs, and 29 chickens. Training data and test data were divided in a total ratio of 8:2.

Model & Hyper Parameter

The experimental setting is as follows:

  • Optimization function : Adam optimizer
  • Learning rate : 0.001, the
  • Batch size : 10, and the
  • Epoch : 100.
  • Sampling rate : 16,000 (MFCC)

Experiment

The loss value of the evaluation set : 0.0324

the accuracy was 279 out of 280