Research Of Measuring Of Appearance Quality Of Rice By Computer Vision

Now,　most　the　measuring　of　appearance　quality　of　rice　in　our　country　still　relies　on　naked　eyes,　which　lacks　objectivity　and　repetitiveness,　so　that　the　rice　with　high　rate　of　chalkiness,　the　broken　grain,　the　crude　grain　and　the　rice　with　ill　speckle　and　mildew　go　into　the　market,　at　the　same　time,　it　reduce　the　competition　of　Chinese　native　rice.　To　compensate　these　weaknesses,　this　research　selected　three　kinds　of　common　rice　and　developed　a　Matlab　assessment　system　of　the　appearance　quality　of　rice　by　employing　computer　vision　technology.　This　system　had　visual　interface　and　the　function　of　man-machine　interaction,　and　can　identify　the　rice's　appearance　property　characters　such　as　chalkiness　and　yellow　rice.　The　main　research　content　as　follows:　1.　Through　analyzing　the　curvature　feature　of　the　connect　rice,　a　method　of　automatically　segmenting　the　connect　rice　is　studied,　which　is　　　the　same　with　all　kinds　of　protruding　skeleton　image;　　2.　Through　analyzing　the　feature　of　chalkiness　of　the　rice,　an　arithmetic　of　checking　chalkiness　has　been　put　forward,　which　can　detect　the　chalk　area　and　the　chalk　rice,　and　calculate　the　parameters　such　as　chalkiness　rate;　3.　Through　comparing　the　chroma　feature　of　normal　rice　and　yellow　rice,　a　method　of　identifying　yellow　rice　is　studied;　4.　Calculating　8　important　configuration　parameter,　they　are　centroid,　area,　perimeter,　eccentricity,　long　axis　,short　axis,　hue　and　saturation.　Experiments　showed　these　parameters　could　identify　three　different　kinds　of　rice　effectively.　5.　　An　arithmetic　of　classification　of　rice　by　BP　net　is　studied　on　the　basis　of　the　features　of　the　rice's　configuration.　The　conclusion　coming　from　comparison　is　the　sorting　result　of　7-7-3　net　is　satisfying.　The　accuracy　of　this　identifying　rate　is　95%.　6.　　A　software　of　the　measuring　of　appearance　quality　of　rice　by　computer　vision　is　developed　on　the　basis　of　Matlab6.5.　This　software　has　friendly　interface　and　can　operate　conveniently,　which　could　realize　the　pretreatment　of　rice　image,　the　segment　of　connect　rice,　the　detection　of　chalkiness　and　yellow　rice　and　the　classification　of　rice　breed　by　BP　net.　The　result　of　research　shows　that　the　system　could　realize　effective　identification　of　the　rice's　configuration　and　has　feasibility　in　automatic　detection　of　the　appearance　quality　of　rice.