Husk, bacterial leaf blight, and weedy rice classification in paddy seeds using imaging techniques

Seed is the foundation in every agricultural product. Paddy normally contains impurities and contaminations such as weedy rice (WR), infected seed, husk, straw, and others after harvesting from field. To produce good quality certified paddy seeds, it must contain minimum impurity and free from WR, insects, disease, and other matters. In current practice, various types of machines were used to separate impurities. Furthermore, it requires seven to 30 days to detect Bacterial Leaf Blight Disease (BLB) symptoms. These methods were not practical and time consuming. Therefore, the aim of this study is to detect impurities using image processing techniques. Husk, BLB, and WR taken from Variant 1 (V1), Variant 2 (V2), and Variant 3 (V1) were studied. Thermal imaging technique was used to detect husk to differentiate between husk and paddy seeds by analysing the changes of heat reflectance between them due to the differences of internal properties. FLIR E60 thermal camera (FLIR System, West Mailing, Kent, United Kingdom) was used to capture thermal images. Heating treatment was applied for 180s, followed by a cooling treatment for 60s. The results show that average mean pixel values of paddy seeds were higher compared with husks due to higher thermal conductivity of paddy seeds and lower thermal conductivity of husks. Mean pixel values at 25s cooling gave a suitable indicator to separate between seeds and husks. The technique can be used to detect husk with 100% success rate for 20% husk and 40% husk, 98.33% for 60% husk and 97.67% for 100% husk, while 94.33% for 100% seeds. Meanwhile, visible imaging was used for BLB and WR classification because there were differences in colour properties, not in heat reflectance. A Samsung NX2000 digital camera (Samsung, South Korea) was used to capture images of paddy seeds, BLB-infected seeds, and WR seeds. Then, an image segmentation and noise removal were applied. In BLB detection, mean pixel values of 12 colour properties – (Red (R), Green (G), Blue (B), Hue (H), Saturation (S), Value (V), Green Leaf Index (GLI), Green-red Vegetation Index (GRVI), Kawashima Index (IKAW), Principal Component Analysis Index (IPCA), Red-green Ratio Index (RGRI) – were extracted and analysed using independent-sample t-test. Statistical results show a reliable difference between BLB-infected seeds and healthy paddy seeds for G, B, S, GRVI, and VARI. The technique can be used to detect BLB-infected seeds with 88.33%, 100.00%, 95.55%, and 96.33% success rate for 20% BLB, 40% BLB, 60% BLB and 100% BLB, respectively. Mean pixel values of these 12 colour properties and two physical properties (area and major axis length) were used to detect WR. Statistical results show a reliable difference between WR and paddy seeds for area, major axis length, GLI, and RGRI. Classification model was developed based on the analysis of the data and results show the average successful detection of 99.25%. In conclusion, the image processing techniques can be used to detect the impurities of paddy seeds caused by husks (using thermal imaging), BLB (using visible imaging), and WR (using visible imaging and physical properties). The proposed image processing approach is more practical and less time consuming compared with the current practice of detection