| Firmness is an important indicator for determining the maturity of fruits and vegetables,which has a significant impact on their quality and storage capability.The firmness classification of fruits and vegetables is of great significance for improving the profitability of the industry.Currently available non-destructive detection equipment for fruit and vegetable firmness is expensive and the classification accuracy can be greatly affected by environmental factors.To bridge this gap,this study proposed a non-destructive firmness evaluation method that focused on tomatoes and nectarines and combined biomimetics technology to design a flexible mechanical gripper based on the fin effect.A visual tactile-based deep learning non-destructive detection method for fruit and vegetable firmness was proposed,which achieved fruit and vegetable firmness classification and grading.The research provided a technical basis for the classification.of fruit and vegetable firmness.The main research contents and results are as follows:(1)The design of a bionic finger structure based on the fin ray effect.Based on the fin ray effect,a bionic finger structure that met the functional requirements of this study was designed and manufactured to construct a low-cost and easy-to-control visual-tactile flexible mechanical gripper system.A finite element simulation analysis model was established,and the results showed that the stiffness of the flexible mechanical gripper changes slightly in the grasping state,making it possible to achieve flexible and safe grasping.Finally,a non-destructive evaluation test was carried out to analyze the respiratory intensity changes of the grasped fruits and vegetables by using statistical methods.The results showed no significant difference in respiratory intensity between the grasping test group and the control group,indicating that the flexible mechanical gripper could achieve non-destructive grasping.(2)The collection and preprocessing of fruits and vegetables visual-tactile data.The firmness data and visual-tactile data of tomatoes and nectarines were collected.The adaptive threshold segmentation and U-Net network semantic segmentation were used to segment the collected visual-tactile data.The results showed that the U-Net network semantic segmentation method could effectively understand the high-level semantic information of image content in the segmentation task of this study and had good robustness.In addition,to address issues such as insufficient data volume and uneven sample firmness distribution,five different data augmentation methods were employed to expand the sample size of the training set and ensure even distribution across different firmness intervals,thereby improving the generalization ability and prediction accuracy of the trained model.(3)The visual-tactile-based non-destructive detection method for fruit and vegetable firmness.A feature extraction network with an attention mechanism added to the residual neural network was designed,and a long short-term memory neural network was used to capture the temporal relationships in the image data.A visual-tactile sequence image firmness prediction network model was constructed,and the firmness prediction results of different datasets and neural network models were compared.The results showed that the preprocessed dataset and the Res Net34＿CBAM-LSTM model achieved best performance for non-destructive firmness classification.For tomato firmness prediction,the coefficient of determination(R~2)of the test set was 0.773,and the root mean square error(RMSE)was 1.702 N.For nectarine firmness prediction,the R~2of the test set was 0.759,and the RMSE was 6.493 N.Finally,the grading accuracy of different firmness level tomatoes and nectarines by using this method was evaluated,with accuracy rates of 92.3%and90.8%,respectively. |
PDF Full Text Request