HCER-ResNet50: A Deep Learning Framework for Accurate Cashew Nut Classification and Grading
TL;DR
The proposed HCER-ResNet50 algorithm by accurately grading the Cashewnuts improves the revenue of the farmers and reduce post-harvest losses.
Abstract
Background: Cashew nuts are healthy and are commonly found in food products, confectionery, and foods, whereas cashew apples, shell derivatives are used in beverages and other industrial products. The rising commercial value of cashew has augmented the requirement of the effective and consistent grading mechanisms in order to safeguard the quality and market value. Methodology: In this paper, an automated cashew nut-grading system, Hybrid Compression and Enhancement based ResNet-50 (HCER-ResNet50), is suggested to overcome the weaknesses of manual and traditional methods of using machine-vision to grade cashew nuts. The structure combines the image compression, image enhancement and deep learning classification to produce reliable and high-quality grading across different imaging conditions. Noise removal, normalization and resizing prior to processing follow a pattern of consistent image quality, a hybrid compression scheme based on DWT, DFT, and DCT combined with other enhancement tools such as thresholding and contrast stretching enhances the clarity and separability of features. Based on this, a convolutional neural network with ResNet-50 is used to categorize cashew nuts into specific grades with a minimum human intervention. Results: In all image processing procedures, the classification accuracy of the HCER-ResNet50 was between 0.9598-0.9720, as the experimental data showed. The PNG Dataset achieve the best accuracy (0.9720) and this goes to show that both the transform and the wavelet-based pre-processing methods improve the performance of the grading. Most of the pre-processing methods including thresholding, DWT approximation, DCT and DFT, contrast enhancement, and DWT horizontal methods used approximately 18 to 22 minutes to train and categorize the model with a steady and computationally efficient performance. Furthermore, the Adam optimizer is used to enhance model parameters in the training stage and contributing to enhance the classification outcome. The proposed HCER-ResNet50 algorithm by accurately grading the Cashewnuts improves the revenue of the farmers and reduce post-harvest losses.