Detecting Coffea Arabica L.Growth Stages Based On Digital Repeat Photography

Coffee is one of the three major beverages and it has important economic and nutritional values,and it is important for yield estimation prediction and field management to observe its growth stages especially several flowering events.To realize the automatic observation of coffee’s growth stages,this paper used high spatial and temporal resolution digital repeat images and climatic data acquired by an automatic observation apparatus in coffee plantation in Lujiangba of Baoshan,Yunnan Province,China.The main work includes:(1)An automatic hierarchical region-based method was proposed to segment coffee flowers in digital images.Firstly,superpixel algorithm and superpixel merging were adopted to obtain semantically significant hierarchical regions.Secondly,the step is to extract the color and texture features for each region.Thirdly,support vector machine(SVM)classifier was trained on these features to recognize the regions belonging to coffee flowers.The result shows our proposed Iab2SPMG method achieves the best performance,and the optimal depression shooting angle for segmenting coffee flowers is 77.5°and best number of merging regions is 500 under that angle achieving intersection over union(IoU)of 0.56.Meanwhile,the amount of segmented flowers correlates significantly with the amount of labeled flowers,and r equals to 0.89.And the regression model of them is significant with the R2 of 0.79.(2)Detecting several flowering events was based on time series flowering regions proportion,then smoothing spline of the flowering regions proportions can be fitted and first derivative curve can be calculated.The dates when regional maximum values of the first derivative curve occur were supposed as flowering days.The testing results show that the time series under azimuth shooting angle of A2 achieves the highest IoU of 0.73 and the time series shooting at 8:00 and 12:00 achieve the best performance of IoU of 0.73 too.(3)Based on several regions of interest,the correlation between the number of flower pixels of five flowering events and the number of fruit pixels of five picking times was tested.It shows that they correlate significantly with the r of 0.57 at the shooting time of 12:00.At the same time,the lagged relationship between flowering and precipitation was explored.The result shows the correlation between the two are 0.33 and 0.35 when flowering lagged behind precipitation by 9 days and 10 days,and they are all extremely significant.In conclusion,our approach can estimate the image-based flower density and detect the coffee sequential flowering events in small fields,so the results can be used for coffee fruit maturity prediction and yield estimation afterwards.