Non-destructive method for assessing fruit quality using modified depthwise separable convolutions on hyperspectral images

Hyperspectral imaging records data over a broad range of electromagnetic spectrum wavelengths and presents a viable option for fruit maturity detection when incorporated with deep neural networks. This paper focuses on improving the accuracy of the Kiwi and Avocado fruit hyperspectral dataset by introducing a modified version of depthwise separable convolution and comparing the results with state-of-the-art models to prove our model's reliability. The research aims to use the proposed model to predict the fruits' ripeness, firmness, and sugar content levels.