Thermal Vision for Lady Finger Quality Assessment: A Three-Stage Convolutional Neural Network Approach

Lady Finger is an important commercial crop that is nutritionally rich and has extensive culinary applications. The present research paper introduced a new technique for the nondestructive postharvest classification of okra pods (Clemson Spineless type) into suitably matured or overmatured okra pods using thermal images and the three-stage CNN technique based on EfficientNetB7. For the thermal photography of lady finger pods, an FLIR E75 Thermal imaging unit was used, and the distance of both the cameras and lady finger was kept at a minimum of more than 0.5 m. The proposed CNN-based technique results were reported in the form of accuracy, sensitivity, specificity, precision, and F1-score, which were 0.9700, 0.9800, 0.9600, 0.9608, and 0.9703, respectively. In other words, the FPR for our classification model was reported to be 0.0400. The results of MCC and kappa statistics were also reported to be 0.9402 and 0.9400, which indicates that the strong agreement was found between the predicted and actual classifications of overmature and suitably matured okra pods. The experiment shows that the suggested approach could be utilized for the nondestructive classification of lady finger pods into suitably matured or overmatured ones.