Estimation Model of Nutritional Content Based on Broiler Feed Images Using Convolutional Neural Network and Random Forest

Abstract

Purpose: This research aims to develop an intelligent model to estimate the nutritional content of broiler chicken feed based on feed images to assist farmers in selecting the best broiler feed and quickly verifying its quality to meet requirements.

Methods: The methodology of this research includes literature study, data collection, data preprocessing, image classification, model evaluation, integration of CNN and random forest models, and estimation of nutritional content based on feed images. We collected 99 samples of broiler chicken feed from online stores in various regions of Indonesia, particularly Java. Next, we took pictures with a smartphone and analyzed the nutritional content using near-infrared spectroscopy. Preprocess the data by enhancing the dataset (color space and data augmentation). We use Convolutional Neural Network (CNN) for the classification of broiler feed images. The performance of the CNN model is evaluated using a confusion matrix. We integrate CNN and Random Forest Regressor (RFR) to estimate nutritional content from the features of broiler feed images.

Result: The performance evaluation shows that the CNN (VGG-16) model is 0.9744% accurate and the RFR model has the highest R2 value of 0.8018. The benefits of this research include faster, more efficient, and automated feed quality measurement compared to traditional methods; maintaining feed quality standards; and avoiding health risks for livestock.

Novelty: This research introduces an intelligent model to estimates the nutritional content of broiler feed images by integrating a CNN model with an RFR.