Soft robots have exhibited excellent compatibility with functional and physical requirements of intraluminal procedures such as bronchoscopy and cardiovascular intervention. Despite their favourable mechanical compliance and scalable design, direct force and shape sensing have proved difficult to be embedded within the soft robot's structure. Also, the rate-dependency of soft sensors requires derivative-based calibration that amplifies data acquisition noise leading to large inaccuracy, especially at small forces. As an alternative method, in this study, we proposed a transfer learning-based calibration schema inherited from GoogLeNet. The proposed method was derivative-free and would capture temporal changes in electrical signals from the soft sensors by capturing image features in scalograms of wavelet transform. WaveLeNet, our derivative-free deep convolutional calibration model, had comparable accuracy over the full range of our soft flexural sensor (<5% error) compared to a previously validated rate-dependent calibration but substantially improved accuracy for small forces (<20mN).