An application of Convolutional Neural Network to lobster grading in the Southern Rock Lobster supply chain

Abstract

Southern Rock Lobster (SRL) is an important commercial export fishery of the Australian economy with a contribution of $250 million annually. However, a range of risks relating to food safety and product fraud requires this industry to develop an effective traceability solution. In response to this biometric identification techniques are seen as a possible solution to provide greater security compared to the current tag-based tracking systems. This paper describes how a Convolutional Neural Network (CNN) can be used in conjunction with image processing techniques to enable an autonomic grading solution in the SRL supply chain. The research is an essential part of an overall investigation into designing a low-cost biometric identification solution for tracking lobsters along their supply chain from catch to table. By using a CNN, the research aims to improve the previous research on lobster grading in establishing a reliable and flexible traceability method to meet different supply chain contexts. In this approach, a pre-trained Mask-RCNN model was adopted to extract regions of interest from lobster images. The deep learning ability of this model allows the carapace areas to be segmented from lobster images automatically for calculating grading attributes including size, weight and colour. This outcome then also generates a high-quality input dataset for the follow-up research on identifying individual lobsters. To prove the effectiveness, the proposed method was validated on a large image dataset collected at a lobster processor and tested on mobile application environment. The findings establish a critical contribution to the complete biometric solution developed for SRL products traceability.