The food industry is in constant demand of performing and easy to implement Non-Destructive Evaluation NDE techniques. In this paper, we tackle the problem of the automatic inspection of fruits and more specifically, the sorting of healthy and damaged fruits, taking apples and peaches as examples. In a recent work, we have explained how to proceed by combining mm-Wave measurements processed with a 2D-FFT and Machine Learning algorithm. The accuracy reaches at least 80%. Although the 2D-FFT is a real-time processing and thus interesting for an industrial implementation, it requires complex measurements, i.e amplitude and phase, which makes the acquisition system more complex. Here we aim to overcome this difficulty by processing amplitude-only measurements. We make use of an image processing based on the direct conversion of the measured amplitude into images. The images form the dataset for the classifier that we choose as a non-linear SVM with a RBF kernel. The advantage of the SVM is that the computational burden is moved to the training phase where we compute the optimal hyper-parameters C* and γ*, while the test is very fast. First, we describe the complete workflow and use a set of apples measured in W-band in Autumn 2019 for validation purpose. We then extend the validation to measurements of peaches conducted over a long time period (summer 2019 and 2020). Finally, we investigate the robustness of the method over frequency while moving to the D-band. For all tests the accuracy ranges from 96% to 100%.