Hybrid quantum neural network models for fruit quality assessment.

Area of Science:

• Quantum Computing
• Machine Learning
• Artificial Intelligence

Background:

• Hybrid quantum neural networks (HQNNs) are emerging as powerful tools for complex pattern recognition tasks.
• Assessing fruit quality often requires sophisticated models capable of handling diverse data characteristics.

Purpose of the Study:

• To investigate the impact of entangling gate choice on HQNN performance for fruit quality assessment.
• To compare two HQNN architectures: NNQEv1 (CNOT gates) and NNQEv2 (CZ gates).

Main Methods:

• Developed two HQNN architectures (NNQEv1 and NNQEv2) using different entangling gates (CNOT vs. CZ).
• Theoretically analyzed architectures based on gate decomposition and hardware noise considerations.
• Computationally executed quantum circuits on classical hardware for performance evaluation.
• Compared HQNN models against classical and deep learning benchmarks.

Main Results:

• Both HQNN models achieved high test accuracies: 98.7% (MNIST), 98.6% (FruitQ), and 96.7% (custom Apple dataset).
• The CZ-based NNQEv2 exhibited more stable training dynamics and tighter cross-validation confidence intervals than the CNOT-based NNQEv1.
• Results supported the theoretical prediction of CZ-gate architecture stability.

Conclusions:

• Gate-level design choices significantly influence HQNN stability and performance.
• The CZ-gate based HQNN architecture is a promising direction for robust quantum machine learning applications.
• This study provides foundational insights for developing advanced quantum machine learning algorithms.