Chili Quality Classification using Convolution Neural Network

Artificial intelligence has propelled transformative advances across industries, particularly in the realm of food quality control, where automation, precise predictions, and informed decision-making have become pivotal. Traditional quality assessment methods, reliant on manual inspection, are resource-intensive, subjective, and time-consuming, necessitating automated and objective alternatives. Convolutional Neural Networks (CNNs) have emerged as potent tools in computer vision, wielding remarkable image analysis capabilities. This study introduces a CNN-based approach to chili quality classification, with the primary aim of creating an automated system capable of categorizing chilies based on various quality attributes. The central objective is to develop a model capable of distinguishing between different quality grades with a high degree of precision. Our findings reveal that MobileNetV2, owing to its inherent model efficiency and adaptability to the small dataset used, excels over InceptionV3. MobileNetV2’s design prioritizes computational efficiency through depthwise separable convolutions, resulting in a more lightweight and compact architecture. In scenarios with limited data, MobileNetV2 demonstrates superior generalization capabilities, striking a balance between model sophistication and dataset scale, as evidenced by its training accuracy of 90.56% and validation accuracy of 86.24%. In contrast, InceptionV3, while proficient with a training accuracy of 92.19%, exhibits comparatively less aptitude for the specific small dataset, with a validation accuracy of 80.6%.