---
license: mit
language:
- en
---
# 🐔 EfficientNetV2S Poultry Feces Classifier

A convolutional neural network model based on **EfficientNetV2S** for classifying chicken fecal images into 4 common conditions:

* **Coccidiosis**
* **Healthy**
* **Newcastle Disease**
* **Salmonella**

This model is designed to support smart poultry farming by enabling early detection of diseases through image-based feces analysis.

## 🧬 Model Architecture

* Base: `EfficientNetV2S` (pretrained on ImageNet, frozen then fine-tuned)
* Head:

  * `GlobalAveragePooling2D`
  * `Dense(128) + BatchNorm + ReLU + Dropout(0.3)`
  * `Dense(4, activation='softmax')`

## 🧪 Training & Evaluation

* Optimizer: Adam
* Loss: Categorical Crossentropy
* Metric: Accuracy
* Dataset:

  * Source: [Jayavrinda et al., 2023](https://doi.org/10.34740/KAGGLE/DS/3951043)
  * 4 classes, resized to 224x224 pixels
  * Train/Val/Test sampling (3k/400/400 per class)
* EarlyStopping was used to monitor validation accuracy
* Accuracy on validation set: **\~90%+** (see notebook for full results)

## 🗄️ Example Usage

```python
from tensorflow.keras.models import load_model
import tensorflow as tf
from PIL import Image
import numpy as np

model = load_model("path/to/your_model.h5")

def preprocess(image_path):
    img = Image.open(image_path).resize((224, 224))
    img_array = np.array(img) / 255.0
    return np.expand_dims(img_array, axis=0)

pred = model.predict(preprocess("feces.jpg"))
class_names = ["Coccidiosis", "Healthy", "Newcastle", "Salmonella"]
print("Prediction:", class_names[np.argmax(pred)])
```

## 📜 Citation

If you use this model or dataset, please cite:

> Jayavrinda Vrindavanam, Pradeep Kumar, Gaurav Kamath, Chandrashekar N, and Govind Patil. (2023). *Poultry Pathology Visual Dataset* \[Data set]. Kaggle. [https://doi.org/10.34740/KAGGLE/DS/3951043](https://doi.org/10.34740/KAGGLE/DS/3951043)

---

Beyond the Outliers

Datathon 2025