Spiking neural networks for apple quality monitoring
Abstract
Sustainable agriculture is a global priority due to its role in food security and economic development. Nevertheless, food waste remains a critical challenge, with UNEP reporting 1.05 billion tons wasted worldwide in 2022, including over 24 million tons in México. Artificial intelligence (AI) and machine vision have advanced agricultural inspection and quality control, yet conventional neural networks often consume significant energy. Neuromorphic computing, inspired by the human brain, offers superior efficiency and parallel processing, making it a promising alternative for fruit quality assessment. A spiking convolutional neural network (SCNN) was developed and trained using the Fruit360 dataset, which includes apples under diverse conditions such as damage, worm infestation, and varying maturity stages. Preprocessing techniques—including Local Binary Pattern (LBP), HSV filtering, and kernel-based feature extraction—were applied to enhance classification accuracy. The system was simulated in Google Colab and tested with the BrainChip Akida AKD1000 coprocessor under Ubuntu 22.04, despite integration challenges with TensorFlow and CUDA libraries. These results highlight neuromorphic computing's potential for high efficiency and low power consumption, addressing practical concerns about energy use in agricultural applications and supporting sustainable practices. Highlight the neuromorphic system's high accuracy of 96% and its low energy consumption of 184.8 mW during training to demonstrate its suitability for sustainable apple quality assessment and energy-efficient applications. Stress that conducting extensive real-world validation and expanding datasets are vital steps to build confidence and motivate collaborative efforts for practical deployment across diverse agricultural settings.