<![CDATA[Computer vision-based assessment of the freshness of fruits and vegetables is imperative for many agricultural applications, including automated harvesting and oversight of the supply chain. This manuscript investigates the advancement of deep convolutional neural networks (CNNs) aimed at detecting fruit freshness through the implementation of multi-task learning (MTL). The proposed models, MTL_ResNet-50 and MTL_CNN, leverage shared feature extraction to concurrently enhance freshness detection and fruit-type classification tasks. The MTL_ResNet-50 and MTL_CNN architectures attained accuracies of 99.32% and 97.95%, respectively. By exceeding the efficacy of established methodologies such as single-task learning and manual feature extraction techniques and by employing an imbalanced dataset to bolster the robustness of our models, we address significant gaps within the existing literature. Furthermore, we markedly surpassed the performance of prior approaches, such as InceptionV3 and CNN_bidirectional long short-term memory (CNN_BiLSTM), resulting in substantial implications for enhancing food quality and safety within both agricultural and consumer domains. These implications encompass the improvement of automated quality control mechanisms within the food industry, the mitigation of food waste, and the assurance of heightened standards for food safety. Future research initiatives may concentrate on augmenting model scalability, refining computational efficiency, and investigating additional applications within agricultural technology.]]>
Copyrights © 2026
