Automated Bruise Detection in Fruits Using AI-Driven Instance Segmentation for Enhanced Quality Control
Disciplines
Manufacturing | Other Mechanical Engineering
Abstract (300 words maximum)
Effective quality control is essential in agriculture to reduce fruit loss, with bruise detection being critical to ensure product integrity from farm to market. This research introduces an advanced AI-driven system that leverages convolutional neural networks (CNNs) to automate bruise detection in fruits. By utilizing Detectron2’s instance segmentation and object detection capabilities with a ResNest-50 backbone, our system accurately identifies bruised areas on fruit surfaces, addressing a key need in post-harvest processes. Bruising in fruits often arises from increased metabolic and physiological changes as they ripen, making them susceptible to damage even from minor external forces. Detecting bruising early can significantly reduce post-harvest losses.
Our method relies on a dataset of 100 annotated high-resolution images, developed with the help of RoboFlow and tailored to capture bruising variations across fruit types like apples and pears. The dataset was divided into training, validation, and testing sets, with 70 images used for training, 20 for validation, and 10 for testing. The model was trained using augmented images to enhance generalization and evaluated on the validation set to tune its parameters. Final testing on a separate test set demonstrated the model's high recall and precision, proving effective in consistently detecting and segmenting bruised regions. This automated approach reduces dependency on manual inspection, minimizes human error, and is scalable for large-scale operations, making it suitable for integration into commercial sorting systems.
In addition to enhancing operational efficiency, this solution reduces labor costs, minimizes waste associated with undetected bruises, and promotes sustainable practices by extending produce shelf life. The proposed solution reflects a forward step in precision agriculture, marking a significant advancement toward data-driven, modernized quality control in the fruit industry.
Academic department under which the project should be listed
SPCEET - Mechanical Engineering
Primary Investigator (PI) Name
Sathish Kumar Gurupatham
Automated Bruise Detection in Fruits Using AI-Driven Instance Segmentation for Enhanced Quality Control
Effective quality control is essential in agriculture to reduce fruit loss, with bruise detection being critical to ensure product integrity from farm to market. This research introduces an advanced AI-driven system that leverages convolutional neural networks (CNNs) to automate bruise detection in fruits. By utilizing Detectron2’s instance segmentation and object detection capabilities with a ResNest-50 backbone, our system accurately identifies bruised areas on fruit surfaces, addressing a key need in post-harvest processes. Bruising in fruits often arises from increased metabolic and physiological changes as they ripen, making them susceptible to damage even from minor external forces. Detecting bruising early can significantly reduce post-harvest losses.
Our method relies on a dataset of 100 annotated high-resolution images, developed with the help of RoboFlow and tailored to capture bruising variations across fruit types like apples and pears. The dataset was divided into training, validation, and testing sets, with 70 images used for training, 20 for validation, and 10 for testing. The model was trained using augmented images to enhance generalization and evaluated on the validation set to tune its parameters. Final testing on a separate test set demonstrated the model's high recall and precision, proving effective in consistently detecting and segmenting bruised regions. This automated approach reduces dependency on manual inspection, minimizes human error, and is scalable for large-scale operations, making it suitable for integration into commercial sorting systems.
In addition to enhancing operational efficiency, this solution reduces labor costs, minimizes waste associated with undetected bruises, and promotes sustainable practices by extending produce shelf life. The proposed solution reflects a forward step in precision agriculture, marking a significant advancement toward data-driven, modernized quality control in the fruit industry.