A Self-Supervised Anomaly Detector of Fruits based on Hyperspectral Imaging ——— A Pytorch Implementation

Requirements

• python==3.x
• torch==1.2.0
• numpy==1.19.5
• scikit-learn==0.20.2
• scikit-image==0.17.2

How to run

• You can now train the SSAD using default parameters using python3 train.py
• In order to get results. you can run the following command python3 test.py

Anomaly detection results of the blueberry data set

you can check result: result.csv

Methods	AUC	F1 Score	Acc_normal	Acc_bruised	Acc_infected	Acc_chilling	Acc_wrinkled
OCSVM	0.744±0.005	0.710±0.005	0.622±0.007	0.789±0.010	0.582±0.013	0.900±0.006	0.614±0.013
AE-1D	0.818±0.028	0.779±0.025	0.712±0.033	0.838±0.044	0.769±0.057	0.743±0.035	0.772±0.027
VAE-1D	0.794±0.008	0.754±0.009	0.678±0.011	0.910±0.014	0.630±0.020	0.823±0.005	0.691±0.013
AE-2D	0.655±0.016	0.643±0.011	0.534±0.015	0.699±0.037	0.991±0.002	0.602±0.031	0.249±0.023
VAE-2D	0.803±0.005	0.768±0.003	0.697±0.004	0.864±0.006	0.963±0.003	0.420±0.006	0.774±0.007
SSAD	0.932±0.015	0.875±0.012	0.837±0.016	0.944±0.018	0.909±0.037	0.838±0.021	0.810±0.027

Anomaly detection results of the strawberry data set

Methods	AUC	F1 Score	Acc_normal	Acc_bruised	Acc_infected	Acc_chilling	Acc_contaminated
OCSVM	0.773±0.009	0.758±0.007	0.643±0.009	0.788±0.020	0.594±0.015	0.904±0.005	0.776±0.003
AE-1D	0.748±0.005	0.727±0.005	0.597±0.007	0.684±0.007	0.552±0.009	0.995±0.001	0.902±0.005
VAE-1D	0.829±0.004	0.784±0.005	0.681±0.008	0.742±0.012	0.496±0.016	0.753±0.018	0.869±0.015
AE-2D	0.690±0.024	0.690±0.017	0.543±0.026	0.460±0.051	0.764±0.013	0.949±0.025	0.866±0.021
VAE-2D	0.659±0.021	0.704±0.014	0.542±0.022	0.475±0.014	0.767±0.021	0.914±0.003	0.717±0.015
SSAD	0.913±0.006	0.869±0.005	0.807±0.007	0.835±0.031	0.834±0.025	0.876±0.022	0.945±0.018

License

MIT © Yisen Liu