Analysing the Quality of Food Using Convolution Neural Network and Fuzzy Classifier in Hyperspectral Imagery
T. Arumuga Maria Devi, Darwin.P, Mebin Jose, P. Kumar- Radiology, Nuclear Medicine and imaging
Introduction:
This study introduces a novel evaluation approach by combining a convolutional neural network (CNN) with a fuzzy neural network (FNN). This approach by utilizing the fuzzy neural network with a few connected layers has the capability to incorporate feature information, thus enhancing the overall functionality of the neural network. In this approach, the CNN generates feature maps (also referred to as outputs) that represent membership values. These feature maps are then input into the fuzzy layers during the training phase. This integration of feature maps and fuzzy layers enables the network to work with both crisp and fuzzy values, due to the additional information generated within the fuzzy set. As a result, an improvement in the classification accuracy of this innovative approach is reported as compared to traditional methods. By utilizing fuzzy neural networks, which can process both crisp and fuzzy values, this approach capitalizes on the expanded information provided by fuzzy sets.
Methods:
In our proposed model, cross-validation tests were conducted. However, the effectiveness of our model relies on having a larger dataset for training sequences. Currently, our dataset is limited in size. During testing, we used a dataset with a greater amount of information, which helps reveal the model's capability to classify objects. This is particularly important when dealing with cases where crucial information is missing.
Results:
The convolution neural network consists of a tuned convolution layer, heuristic activation operation, and parallel element merge layer, which is manipulated by the fuzzy classifier output based on the food image context extractor.
Conclusion:
In this study, the food quality was analyzed through visual IDE. Furthermore, the hyperspectral output image was also extracted with good accuracy.