A Deep Learning Based Method for Grading the Appearance Quality of Peaches

作者：Chun Kong, Xin Zhou, Shiyao Chen, Feng Feng, Chen Wang, Chunyue Ma, Xiuru Guo, Bo Sun, Peng Liu, Zhijun Wang · 年份：2023 · DOI：10.1109/iaecst60924.2023.10503338 · 研究领域：Spectroscopy and Chemometric Analyses、Smart Agriculture and AI、Leaf Properties and Growth Measurement

To accurately and swiftly grade peach appearance without causing damage, a quality grading method for peach appearance called CS-MobileNet-P is proposed. The method utilizes deep learning convolutional neural network MobileNet V3. A high-quality image database that evaluates the visual quality of peaches is generated by filtering and refining the dataset. Efficiency of the model is increased by incorporating the PReLu activation function and CA attention mechanism module. To enhance the accuracy of the Bneck structure within the model, optimization techniques are employed, alongside the utilization of the learning rate preheating strategy that results in shorter model training time, faster convergence, and increased model generalization ability. The variation in the number of parameters and computational workload is minimal. Following the utilization of a learning rate warm-up strategy during the model training, recognition accuracy of the CS-MobileNet-P model increased to 98.87%. The upgraded CS-MobileNet-P model is capable of reliably and swiftly performing non-invasive peach appearance quality grading.