ftirnet 0.1.2

Deep Learning Based FTIR Polymer Classification

FTIRNet

FTIRNet is a deep learning-powered Python package for polymer identification from FTIR (Fourier Transform Infrared Spectroscopy) spectra.

Using a 1D Convolutional Neural Network (CNN), FTIRNet can classify FTIR spectra into common polymer categories with a simple Python API or command-line interface.

Features

• Deep Learning-based FTIR classification
• Simple Python API
• Command Line Interface (CLI)
• Supports multiple trained models
• Returns prediction confidence scores
• Returns probabilities for all supported polymer classes
• Lightweight package with bundled pretrained weights

Supported Polymer Classes

• HDPE (High-Density Polyethylene)
• LDPE (Low-Density Polyethylene)
• PET (Polyethylene Terephthalate)
• PP (Polypropylene)
• PS (Polystyrene)
• PVC (Polyvinyl Chloride)

---

Installation

pip install ftirnet

---

Input Format

FTIRNet expects a processed FTIR CSV file where:

• Each row represents a spectrum

• Feature columns contain absorbance/transmittance values

• Optional columns:

  • Sample_ID
  • Polymer

Example:

Sample_ID,Polymer,f_600,f_602,f_604,f_606,...
HDPE004,HDPE,95.74,95.69,95.75,95.89,...

For inference, the Polymer column is optional and will be ignored if present.

---

Python API

Basic Prediction

import ftir

result = ftir.predict(
    "sample.csv"
)

print(result.prediction)
print(result.confidence)

Using a Specific Model

import ftir

result = ftir.predict(
    "sample.csv",
    model="pretrained"
)

import ftir

result = ftir.predict(
    "sample.csv",
    model="base"
)

---

Prediction Result Object

The predict() function returns an FTIRResult object.

Prediction

print(result.prediction)

Example:

HDPE

Confidence Score

print(result.confidence)

Example:

93.95

Sample ID

print(result.sample_id)

Example:

HDPE004

Class Probabilities

print(result.probabilities)

Example:

{
    "HDPE": 93.95,
    "LDPE": 2.67,
    "PET": 0.99,
    "PP": 0.55,
    "PS": 1.28,
    "PVC": 0.56
}

Access Individual Class Probabilities

print(result.probabilities["HDPE"])

---

Command Line Interface

Default Prediction

ftir sample.csv

Use Pretrained Model

ftir sample.csv --model pretrained

Use Base Model

ftir sample.csv --model base

Help

ftir --help

---

Available Models

Model	Description
pretrained	Transfer learning based model with higher accuracy
base	Base CNN model trained from scratch

---

Example Output

Sample ID : HDPE004
Prediction: HDPE
Confidence: 93.95%

Probabilities:

HDPE      : 93.95%
LDPE      : 2.67%
PET       : 0.99%
PP        : 0.55%
PS        : 1.28%
PVC       : 0.56%

---

Requirements

• Python 3.11+
• PyTorch
• NumPy
• Pandas
• Scikit-learn
• Joblib

---

Citation

If you use FTIRNet in academic research, please cite the associated project or publication when available.

---

License

MIT License