Computer vision system for characterization of Canadian pulses

加拿大是世界第二大豆类生产国。2016年，中国向150多个国家出口了约600万吨豆类(占总产量的80%以上)，价值42亿美元。目前，在国际PULSE贸易中，大约十分之三的合同导致买卖双方需要仲裁的纠纷，这主要是由于质量标准的方法和术语上的差异。已要求种植和出口脉搏的国家开发衡量和界定其质量的创新方法，以加强认证制度。根据Pulse Canada的说法，向国际市场提供加拿大品牌的豆类并保持竞争地位是一项持续的挑战。进口国给予质量参数的值差异很大，在分级、包装和认证方面严重影响了加拿大农民和贸易商。因此，迫切需要一个客观的系统来测量豆类的表面和内部属性，并根据客户规格确定等级，以保持加拿大在国际豆类市场的份额。计算机视觉是一种非破坏性的方法，通过对物体(静止或运动)的图像进行分析，从而客观地获得目标的质量。在这项拟议的计划中，将开发CV技术来表征加拿大谷物委员会定义的加拿大顶级三种豆类(扁豆、豌豆干和干豆)的主要品质决定因素(损害、坚固程度、异物和其他脉冲类别的混合)。将开发和测试用于单色相机、红绿蓝(RGB)彩色相机和近红外(NIR)相机的CV系统，该系统具有优化的照明、图像处理、特征提取、机器学习算法和分类模型。开发的算法将用于实现加拿大脉搏处理设施的自动在线评分系统。这将有助于确保“加拿大脉搏保证”印章在国际市场上的有效性。此外，开发的方案将有益地补充其他正在进行的加拿大豆类研究项目，如那些专注于储存(颜色退化)和育种(品种纯度的模式识别)的项目。能力建设和知识传播：将对10名HQP(1名PDF、1名博士、3名硕士和5名UG)进行关于食品安全和质量的自动化、无损检验程序的培训。在这一领域接受的培训将可用于涉及其他食品的类似研究或工业应用。将为加拿大豆类部门的农民、加工商和贸易商组织一次讲习班，以传播已开发的技术。

Canada is the second largest producer of pulses in the world. In 2016, it exported around 6 million tonnes of pulses (more than 80% of production) valued at $ 4.2 billion to over 150 countries. At present, in the international pulse trade, around 3 out of 10 contracts results in disputes requiring arbitration between buyers and sellers, which is mainly due to differences in methodology and nomenclature in quality standards. The countries that grow and export pulse have been asked to develop innovative methods to measure and define its quality in order to strengthen certification systems. According to Pulse Canada, delivering "Canada brand" pulses to the international market and maintaining a competitive position is an ongoing challenge. The widely varying values given to the quality parameters by importing countries severely affects Canadian farmers and traders in terms of grading, packing and certification. Therefore, an objective system for measuring the surface and internal attributes of pulses, and determining the grades based on customer specifications, is urgently required to maintain Canada's share in the international pulse market. Computer vision (CV) is a non-destructive method in which images of an object (static or moving) are taken and analyzed to obtain the target quality objectively. In this proposed program, the CV techniques will be developed to characterize the primary quality determinants as defined by the Canadian Grain Commission (damage, degree of soundness, foreign material and mix of other pulse classes) for the top three pulses (lentils, dry peas and dry beans) in Canada. The CV system for the monochrome camera, red-green-blue (RGB) color camera, and near infrared (NIR) camera with optimized illumination, image processing, feature extraction, machine learning algorithms and classification models will be developed and tested. The developed algorithms will be used to implement an automated online grading system for the pulse handling facilities of Canada. This will assist in ensuring the validity of the "Guaranteed Canadian Pulse" stamping in the international market. Also, the developed protocols will beneficially supplement other on-going Canadian pulse research programs such as those focused on storage (color degradation) and breeding (pattern recognitions in varietal purity). Capacity building and knowledge dissemination: Ten HQP (1 PDF, 1 PhD, 3 MSc and 5 UG) will be trained in automated, non-destructive inspection procedures for food safety and quality. The training received in this area will be transferable to similar research or industrial applications involving other food products. A workshop will be organized for the farmers, processors and traders in the Canadian pulse sector, in order to disseminate the developed technologies.