Search

The study extracts representative features to train a model with supervised machine learning (ML) to detect powdery mildew (Sphaerotheca macularis f. sp. fragariae) on the strawberry leaves. Powdery mildew (PM) is a fungal disease that greatly affects the production of strawberry and usually infects under conditions of warming temperatures and high humidity. In this research, we report robust models to detect PM using image processing and ML technologies. Three feature extraction techniques (histogram of oriented gradients; HOG, speeded-up robust features; SURF, and gray level co-occurrence matrix; GLCM) and two supervised ML (artificial neural network; ANN and support vector machine; SVM) were implemented using MATLAB. Images were augmented to 1016 images using a four different angle rotation technique to simulate strawberry leaf bundles in the real field. The classification accuracy (CA) to detect PM was highest at 94.34% with a combination of ANN and SURF with 908 × 908 image resolution and with SVM and GLCM at 88.98% with 908 × 908 image resolution. In terms of the extraction time for real-time processing, HOG takes the shortest time to extract features in both ANN and SVM.

The wild blueberry crop requires the heavy application of agrochemicals for the proper crop yield and weed control. An integrated machine vision based weed detection system was developed to target goldenrod weed spot-specifically. Color co-occurrence matrices and statistical classifiers were used for the goldenrod detection. The linear and quadratic classifiers were developed using different reduced sub-sets of textural features. The classifiers were evaluated using accuracy, specificity, sensitivity, and false negative ratio at Laboratory scale. Performance of the developed weed detection system was also evaluated in two wild blueberry fields. The results indicated that quadratic classifier with DM-HSISD model showed the best performance at laboratory scale with the classification accuracies of 94.98% and 93.80% for training and testing datasets, respectively. Another quadratic classifier (with DM-HSI model) containing all 39 textural features was found to be the second best with the accuracies of 94.29% and 91.46% on training and test datasets, respectively. The linear classifiers didn’t perform well compare to their respective quadratic counter-parts. Field performance of the developed system equipped with the quadratic DM-HSISD classifier indicated no significant difference between variable rate (VR) and uniform application (UA) in terms of mean percentage area coverage (PAC) for the targeted goldenrod spots in both fields. However, a significant difference was observed between mean PAC of VR and UA applications for the non-targeted wild blueberry spots. The potential and actual chemical savings were in ranges between 46.71% and 74.83% and 30.12% and 60.58% depending on the weed and sprayed area, respectively. These results demonstrated that the developed weed detection system has potential for targeted application of agrochemicals to control goldenrod in wild blueberry fields.

With being rapid increasing population in worldwide, the need for satisfactory level of crop production with decreased amount of agricultural lands. Machine vision would ensure the increase of crop production by using an automated, non-destructive and cost-effective technique. In last few years, remarkable results have been achieved in different sectors of agriculture. These achievements are integrated with machine learning techniques on machine vision approach that cope with colour, shape, texture and spectral analysis from the image of objects. Despite having many applications of different machine learning techniques, this review only described the statistical machine learning technologies with machine vision systems in agriculture due to broad area of machine learning applications. Two types of statistical machine learning techniques such as supervised and unsupervised learning have been utilized for agriculture. This paper comprehensively surveyed current application of statistical machine learning techniques in machine vision systems, analyses each technique potential for specific application and represents an overview of instructive examples in different agricultural areas. Suggestions of specific statistical machine learning technique for specific purpose and limitations of each technique are also given. Future trends of statistical machine learning technology applications are discussed.

Wild blueberry crop yields are dependent on heavy agrochemical applications to control weeds competing with the crop. Goldenrod is a creeping herbaceous perennial weed that occurred in 90% of wild blueberry fields surveyed in Nova Scotia. A colour co-occurrence matrices (CCMs) based algorithm can be used for spot-application of herbicide on goldenrod within wild blueberry fields. The objective of this study was to analyse the effect of different parameters on computational complexity and goldenrod identification accuracy of CCMs in wild blueberry cropping system. An image acquisition graphical user interface (GUI) and CCMs based textural analysis algorithm were developed in Microsoft Visual® C# programming language. The GUI was used to acquire 2244 area of interest images along with a set of 2244 full frame images from two different wild blueberry fields in central Nova Scotia, Canada. Six image intensity levels, seven image sizes, and three CCMs were used to for the study. The results indicated that intensity levels and image size significantly influenced the computational requirements of CCMs. Images with 256 or 128 intensity levels could be used as these levels correctly classified 94% and 89% of test observations respectively. The processing times were increased from 535 μs to 10,650 μs and 9864 μs to 63,750 μs as the intensity increased from 8 to 256 levels and image size increased from 16 × 16 to 1024 × 1024 pixels, respectively. The time required for textural feature extraction was not statistically significant for different image sizes used in this study. Overall, the results indicated that intensity levels of 128 or 256, a unit image size of 128 × 128 pixels, and saturation colour plane alone or in combination with hue can help to minimise the processing burden without compromising the classification accuracy for real-time applications.

Spatial variations in wild blueberry plant height pose a serious challenge for the operator to maintain the optimum harvester picking head height. Harvester head adjustment based on plant height increases harvestable fruit yield and quality while preventing plant pulling. An advanced Ultrasonic Plant Height Measurement System II (UPHMS II) was developed and incorporated into a commercial mechanical harvester. The developed system consisted of three ultrasonic sensors covering the width of the harvester head, a real-time kinematics global positioning system (RTK-GPS), custom built software, and a rugged computer. The custom software acquired and processed the ultrasonic sensing data in real-time during mechanical harvesting. Four wild blueberry fields were selected in central Nova Scotia to evaluate the performance of the developed system. Forty eight experimental plots were randomly constructed within four fields and wild blueberry plant heights were recorded manually prior to harvest. The UPHMS II was tested and evaluated to estimate plant height. The manual plant height measurements were compared with ultrasonically sensed data to ensure the accuracy of the developed system. A previous system (UPHMS I) comprising of one ultrasonic sensor was also tested and evaluated. Results of regression and scatter plots revealed that the UPHMS II was able to sense plant height in real-time more accurately with root mean square error (RMSE) of 1.7 cm when compared with the UPHMS I (RMSE = 5.7 cm). The UPHMS II equipped with three sensors covered the complete width of the harvester (0.91 m), which showed higher accuracy compared to the UPHMS I.

The goal of the project was to supply growers with knowledge on how incorporation of machine vision technology can affect the wild blueberry crop, disease pressures, and the overall savings of select agrochemical inputs. A machine vision system was developed and mounted on a rear sprayer boom in front of the sprayer nozzles capable of targeting the agrochemical application on an as-needed basis. Results showed that plants that received the proper fungicide application were less prone to premature leaf drop resulting in larger stem diameters and higher bud counts and harvestable fruit yield. Fungicide application savings using the smart sprayer for spot-application was 12% as compared to a uniform application.

In this study, Deep Learning (DL) was used to detect powdery mildew (PM), persistent fungal disease in strawberries to reduce the amount of unnecessary fungicide use, and the need for field scouts. This study optimised and evaluated several well-established learners, including AlexNet, SqueezeNet, GoogLeNet, ResNet-50, SqueezeNet-MOD1, and SqueezeNet-MOD2. Data augmentation was carried out from among 1450 healthy and infected leaf images to prevent overfitting and to consider the various shapes and direction of the leaves in the field. A total of eight clockwise rotations (0°; the original data, 45°, 90°, 135°, 180°, 225°, 270°, and 315°) was performed to generate 11,600 data points. Overall, the six DL algorithms that were used in this study showed on average of >92% in classification accuracy (CA). ResNet-50 gave the highest CA of 98.11% in classifying the healthy and infected leaves; however, considering the computation time, AlexNet had the fastest processing time, at 40.73 s, to process 2320 images with a CA of 95.59%.When considering the memory requirements for hardware deployment, SqueezeNet-MOD2 would be recommended for PM detection on strawberry leaves with a CA of 92.61%.

The present study assessed the ability of a low cost dual-view computer-vision system (CVS) to measure volume and color co-occurrence image textural features of apple slices to find the end of drying process by comparing physical texture parameters and moisture content. Apple slices were dried at three different temperatures (40, 60 and 80 °C) and their volumes were measured with both the CVS and a caliper. The physical texture parameters and image textural features were measured according to the moisture content. While the CVS can be used to measure the volume of the apple slices during the drying process, the volume was not a good indicator of the end of drying because of porosity development. The peak force and initial slope of physical texture showed a significant correlation with H3, H5 and S2 and H3, H5 and S11 of image texture features, respectively. Eleven image texture features correlate well with the moisture content ( R 2 > 0.9) and also had a high correlation between each other. The uniformity of intensity (I1) of image feature was a good indicator of the end of drying because of the close correlations with moisture content. The drying times according to the moisture content and the uniformity of intensity were 4.31 ± 0.313 h and 4.42 ± 0.125 h, respectively.

Sensing and control system was developed for spot-application of fertilizer in real-time.Software was made to process data to differentiate foliage and bare spots.The modified fertilizer spreader was efficient and accurate enough for spot-application of fertilizer.Spot-application can increase profitability and mitigate the environmental risks. The commercial variable rate (VR) fertilizer spreader can change fertilizer rates in different management zones using global positioning system (GPS) guided prescription maps. The VR fertilizer spreader allows the independent control only on half of the fertilizer boom (3.66m; 6 nozzles). This existing VR fertilizer spreader fertilizes unevenly distributed small sized bare spots/weed patches which are less than half boom length in wild blueberry fields. Therefore, the existing VR fertilizer spreader was modified to control each pair of nozzles for spot-application of fertilizer only in plant areas. The automated sensing and control system consisting of 6µEye color cameras, solenoid valves, pneumatic cylinders, programmable logic controller (PLC), 8-channel VR controller (VRC) interfaced to a pocket PC (PPC) using wireless Bluetooth? radio with Windows Mobile? and a custom software was developed. The µEye color cameras were mounted on a separate boom in front of the tractor at a height of 1.2m. The modified VR spreader was capable of using prescription maps and automated sensing and control system simultaneously to detect foliage/bare spots in real-time to avoid fertilization in bare spots/weed patches. The ACCU-RATE controller calibration and response time calculations were performed prior to field experiment.The results of calibration tests for the ACCU-RATE controller suggested 0.05) between VR and UN treatments in grass/plant areas. These results suggested that the MVRG fertilizer spreader was efficient and accurate for spot-application of fertilizer to increase farm profitability and reduce environmental risks.

Site-specific crop management can increase profitability and reduce environmental risks of wild blueberry crop having large spatial variation in soil/plant characteristics and topographic features which may affect fruit yield. An integrated automated system comprising of an ultrasonic sensor, a digital color camera, a slope sensor, a real-time kinematics GPS (RTK-GPS), custom software and ruggedized computer was developed. The system was incorporated onto a commercial wild blueberry harvester, to measure plant height, fruit yield, slope and elevation simultaneously while harvesting. Two wild blueberry fields were selected in central Nova Scotia, Canada to evaluate the performance of the system. Field boundaries, bare spots, weeds and grass patches were mapped with a RTK-GPS prior to start the experiment. Linear regression was used to calibrate the actual fruit yield with the percentage of blue pixels (R^2=0.91-0.92; P

Similar to most biological studies, beef contamination classification studies using artificial neural networks are restricted to small datasets. This study evaluates multivariate normal (MVN) technique of synthetic sample generation on small datasets associated with Salmonella contamination in beef. Six experiments were conducted to evaluate the performance of integrated sensor system towards identification of Salmonella contaminated beef packages. Pattern recognition involved using wavelet packet transform for feature extraction from sensor array responses and radial basis function network (RBFN) based classification of contaminated beef packages from uncontaminated packages. The MVN generated synthetic olfactory sensor signatures were used to train and test the RBFN classifiers. For the datasets analyzed in this study, genetic algorithm optimized RBF networks conferred average contamination test classification accuracies of 90.33% ± 7.68% (mean ± std. dev.) which were higher compared to the bootstrapped quadratic discriminant analysis based average accuracies. RBFN classifier based average overall classification accuracies of six synthetically generated datasets were in the range of 86.66%–98.89% with highest average overall classification accuracies of 98.89% ± 1.92%.

Wild blueberry fruit yield maps could be used to generate prescription maps for site-specific application of fertilizer to improve crop productivity. An automated yield monitoring system II (AYMS II) consisting of two @mEye color cameras, real time kinematics-global positioning system, custom software, and a ruggedized laptop computer was developed and mounted on a Specialized Farm Motorized Vehicle (SFMV) for real-time fruit yield mapping. Custom software was developed in C++ programming language to acquire and process image in real-time from both cameras and store the percentage of blue pixels in ruggedized laptop computer. Two wild blueberry fields were selected in central Nova Scotia to evaluate the performance of the AYMS II for fruit yield mapping. Calibration was carried out at 20 randomly selected data points in each field. The ripe fruit was hand-harvested out of a 2.0x0.7m quadrat at each selected point and camera images were also taken from the same points to calculate the percentage of blue pixels (fraction of blue pixels in the image). Linear regression analysis was performed to calibrate the actual fruit yield with the percentage of blue pixels. Real-time yield mapping was carried out with AYMS II. The estimated yield along with geo-referenced coordinates was imported into ArcGIS 9.3 software for detailed mapping. The AYMS II, hardware and software performed well to estimate wild blueberry fruit yield. The linear regression results showed highly significant relationship between the percentage of blue pixels and actual fruit yield in field 1 and field 2. The correlation between actual and predicted fruit yield (validation, using the equation from field 2) in field 1 and field 2 (validation, using the equation from field 1) was also highly significant. Maps developed in ArcGIS 9.3 showed substantial variability in fruit yield in both fields. The bare spots were coincided with no or low yielding areas in the fields. The viable AYMS II will be incorporated into wild blueberry harvester. This information obtained from AYMS II could be used to implement site-specific management practices within the blueberry fields to optimize productivity while minimizing the environmental impact of farming operations.

An automated prototype variable rate (VR) sprayer was developed for control of 8 individual nozzles on a 6.1m sprayer boom for in-season, site-specific application of agrochemicals on weeds. The sprayer boom was divided into 8 sections and mounted behind an all-terrain vehicle (ATV) at 76.2cm above the ground. The variable-rate control system consisted of 8 ultrasonic sensors (one per spray section) mounted on a separate boom in front of the ATV, DICKEY-john Land Manager II controller and flow valve, solenoid valves and an 8-channel variable rate controller interfaced to a Pocket PC (PPC) using wireless Bluetooth^(R) radio with Windows Mobile^(R) compatible software. This type of VR sprayer does not use prescription maps, but relies on sensors to provide real-time weed detection information which is used to dispense correct agrochemical rates for the weeds. The sprayer can be used for in-season, spot application (SA) of agrochemicals by activating specific boom sections where the weeds have been detected. Two wild blueberry fields have been selected in central Nova Scotia to evaluate the accuracy of the VR sprayer. Water sensitive papers (targets) were stapled to weeds randomly selected in two tracks of each field. The papers were orientated parallel to the ground. The percent area coverage (PAC) of the sprayed targets with both SA and uniform application was calculated by an imaging system. Non significance of the t-test for uniform versus SA targets PAC indicated that there was no significant bias in the SA and that the SA was accurate. The PAC with both applications ranged from 10.01% to 81.22% and from 5.39% to 72.67% in field 1 and field 2, respectively. Weed heights at selected points were measured and related with percent area coverage to examine the sprayer performance for spray application. Linear regression analysis showed that weed heights were significantly correlated (R^2 ranged from 0.60 to 0.75) with percent area coverage of the targets in selected fields with both applications. It is proposed that herbicide should be applied at early stage of weed growth (weed height ranged from 35cm to 55cm and plant height ranged from 12cm to 30cm) for appropriate application with these specific VR sprayer arrangements. Based on these results, the VR sprayer was efficient and accurate enough for spot-application of agrochemicals usage in wild blueberry fields.