Identifying cherry maturity and disease using different fusions of deep features and classifiers

Yang, Jing; Wang, Guo Hui

doi:10.1007/s11694-023-02091-4

Cited by 1 publication

(1 citation statement)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Concurrently, this has also introduced novel solutions for crop identification. As a result, nowadays, the control and monitoring of fruit appearance quality by electronic ways such as machine vision and deep learning has been increasingly taking the place of manual means in some developed countries [8][9][10][11]. Compared with the manual detection of cherry appearance quality, the advantages of machine vision and deep learning techniques include high accuracy and detection speed, high flexibility and low costs, program-mability.…”

Section: Introductionmentioning

confidence: 99%

A Swin transformer and MLP based method for identifying cherry ripeness and decay

Song,

Yang,

Wang

2023

Front. Phys.

Self Cite

View full text Add to dashboard Cite

Cherries are a nutritionally beneficial and economically significant crop, with fruit ripeness and decay (rot or rupture) being critical indicators in the cherry sorting process. Therefore, accurately identifying the maturity and decay of cherries is crucial in cherry processing. With advancements in artificial intelligence technology, many studies have utilized photographs for non-destructive detection of fruit appearance quality. This paper proposes a cherry appearance quality identification method based on the Swin Transformer, which utilizes the Swin Transformer to extract cherry image feature information and then imports the feature information into classifiers such as multi-layer perceptron(MLP) and support vector machine(SVM) for classification. Through the comparison of multiple classifiers, the optimal classifier, namely, MLP, in combination with the Swin Transformer is obtained. Furthermore, performance comparisons are conducted with the original Swin-T method, traditional CNN models, and traditional CNN models combined with MLP. The results demonstrate the following: 1) The proposed method based on the Swin Transformer and MLP achieves an accuracy rate of 98.5%, which is 2.1% higher than the original Swin-T model and 1.0% higher than the best-performing combination of traditional CNN model and MLP. 2) The training time required for the Swin Transformer and MLP is only 78.43 s, significantly faster than other models. The experimental results indicate that the innovative approach of combining the Swin Transformer and MLP shows excellent performance in identifying cherry ripeness and decay. The successful application of this method provides a new solution for determining cherry appearance ripeness and decay. Therefore, this method plays a significant role in promoting the development of cherry sorting machines.

show abstract

Section: Introductionmentioning

confidence: 99%