Location of seed spoilage in mango fruit using X-ray imaging and convolutional neural networks

Abstract

Detection of the internal damage and insect infestation on intact mango fruit during harvesting, storage, and exportation is particularly challenging. Visual comparison of infestation by spot detection has proven unreliable in determining the presence of insects in seeds and the extent of spoilage in the fruit. The study aimed toward identifying a non-destructive, rapid, and correct method for detecting spoilage in mango seeds. Ninety-eight (98) mature green mango fruits were harvested from a farm (Kintampo, Ghana) for the experiment. The mangoes were numbered for photographic and X-ray imaging. The images were visually examined and classified based on morphology into mango fruit with good seed and bad seed. The dataset was then augmented into 8000 images and split into train (4000) and test (4000) sets. A VGG16 Deep Convolutional Neural Network (DCNN) model was trained, tested, cross-validated (five-fold), and evaluated by saliency maps on visualization algorithm, Grad-weighted Class Activation Mapping, Grad-CAM, and the Confusion matrix to ascertain its capacity to identify bad seed from good seed. The model achieved an accuracy rate of 97.66% and 0.988 area under the receiver operations characteristics curve. Further application of this model in the mango industry could promote non-destructive early detection of seed deterioration on the field and enhance the quality of products throughout the postharvest supply chain.